heavy-duty/ceremony

Fork 2

fix: facts.sh reads a parentless head as base_ver=(none), not exit 128 #136

Merged

claude-bot-andresmgsl merged 3 commits from build/134-facts-root-commit into main

2026-07-24 12:47:34 +00:00

claude-bot-andresmgsl commented

2026-07-24 12:14:30 +00:00

(Migrated from github.com)

Closes #134.

A repository's first push to main is a branch-create push (event.before all-zeros) whose head is a root commit — MERGE_SHA^1 does not exist, rev-parse fails, and set -e killed lib/facts.sh at exit 128 before a single fact was established. Both 0.2.0 drills hit it independently. Per the issue's decided spec: the head's parent count is read as a fact (git rev-list --parents -n 1, D2), a zero-parent head takes the existing base_ver="(none)" path (D1) with the belt-and-braces fetch and the base git show skipped so no rev-scoped command ever runs with an empty rev (D3), and the comments carry the 0.2.0-drill war story (D4). An unresolvable MERGE_SHA still dies loudly — the parents= assignment propagates rev-list's 128 under set -e, never masquerading as (none).

Acceptance criteria (from #134)

facts.sh with EVENT_BEFORE all-zeros against a root commit exits 0 and prints base_ver=(none) — test/facts.test.sh "root commit, all-zeros event.before"
Same, with EVENT_BEFORE empty (the non-push caller) — "root commit, empty event.before"
A root commit whose VERSION is 0.1.0-dev produces ceremony=no and a green NOTICE through the full facts→decide chain — test/release-chain.test.sh greenfield cases
A root commit whose VERSION is bare 0.1.0 with no merged release-labeled PR still refuses (exit 1) and creates nothing — release-chain "a bare unlabeled root commit still refuses"
An unresolvable MERGE_SHA still fails loudly (exit 128) and is not reported as base_ver=(none) — the D2 pin, both the positive (dies with "bad object") and the negative (no (none) in output) assertions
No git show, git cat-file or git fetch runs with an empty rev on the no-base path — both are behind [ -n "$base_sha" ]; the greenfield fixtures run with the refuse-by-default gh stub and no origin, so a stray fetch/show would surface
Every existing test/facts.test.sh case passes unchanged — including L96–L99, all-zeros/empty event.before on a parented head still resolve to the first parent
test/run.sh (17 files), CI shellcheck sweep (shellcheck -x, 37 scripts), actionlint — all green locally

Worklog

lib/facts.sh: parent count read via rev-list --parents; zero parents → (none) path; fetch + base show skipped on no-base; comments name the 0.2.0 drills and #134 (D1–D4)
test/facts.test.sh: greenfield fixtures (zeros + empty event.before, bare root with labeled-no), D2 pin (bad MERGE_SHA → 128, no (none) in output)
test/release-chain.test.sh: end-to-end — greenfield -dev root goes green ceremony=no with a NOTICE; bare unlabeled root refuses exit 1
changelog.d/134.md
Full suite + shellcheck + actionlint green locally
Mark ready, request the panel — codex, grok, kimi requested

Closes #134. A repository's first push to `main` is a branch-create push (`event.before` all-zeros) whose head is a root commit — `MERGE_SHA^1` does not exist, `rev-parse` fails, and `set -e` killed `lib/facts.sh` at exit 128 before a single fact was established. Both 0.2.0 drills hit it independently. Per the issue's decided spec: the head's parent count is read as a fact (`git rev-list --parents -n 1`, D2), a zero-parent head takes the existing `base_ver="(none)"` path (D1) with the belt-and-braces fetch and the base `git show` skipped so no rev-scoped command ever runs with an empty rev (D3), and the comments carry the 0.2.0-drill war story (D4). An unresolvable `MERGE_SHA` still dies loudly — the `parents=` assignment propagates `rev-list`'s 128 under `set -e`, never masquerading as `(none)`. ## Acceptance criteria (from #134) - [x] `facts.sh` with `EVENT_BEFORE` all-zeros against a root commit exits `0` and prints `base_ver=(none)` — `test/facts.test.sh` "root commit, all-zeros event.before" - [x] Same, with `EVENT_BEFORE` empty (the non-push caller) — "root commit, empty event.before" - [x] A root commit whose `VERSION` is `0.1.0-dev` produces `ceremony=no` and a green NOTICE through the full facts→decide chain — `test/release-chain.test.sh` greenfield cases - [x] A root commit whose `VERSION` is bare `0.1.0` with no merged release-labeled PR still refuses (exit 1) and creates nothing — release-chain "a bare unlabeled root commit still refuses" - [x] An unresolvable `MERGE_SHA` still fails loudly (exit 128) and is not reported as `base_ver=(none)` — the D2 pin, both the positive (dies with "bad object") and the negative (no `(none)` in output) assertions - [x] No `git show`, `git cat-file` or `git fetch` runs with an empty rev on the no-base path — both are behind `[ -n "$base_sha" ]`; the greenfield fixtures run with the refuse-by-default gh stub and no `origin`, so a stray fetch/show would surface - [x] Every existing `test/facts.test.sh` case passes unchanged — including L96–L99, all-zeros/empty `event.before` on a parented head still resolve to the first parent - [x] `test/run.sh` (17 files), CI shellcheck sweep (`shellcheck -x`, 37 scripts), actionlint — all green locally ## Worklog - [x] `lib/facts.sh`: parent count read via `rev-list --parents`; zero parents → `(none)` path; fetch + base show skipped on no-base; comments name the 0.2.0 drills and #134 (D1–D4) - [x] `test/facts.test.sh`: greenfield fixtures (zeros + empty `event.before`, bare root with `labeled-no`), D2 pin (bad `MERGE_SHA` → 128, no `(none)` in output) - [x] `test/release-chain.test.sh`: end-to-end — greenfield `-dev` root goes green `ceremony=no` with a NOTICE; bare unlabeled root refuses exit 1 - [x] `changelog.d/134.md` - [x] Full suite + shellcheck + actionlint green locally - [x] Mark ready, request the panel — codex, grok, kimi requested

danmt (Migrated from github.com) reviewed 2026-07-24 12:14:30 +00:00

codex-bot-andresmgsl commented

2026-07-24 12:20:41 +00:00

(Migrated from github.com)

🔎 reviewing head fa768510bf

🔎 reviewing head fa768510bff3624f40716b53c115968271015ae1

grok-bot-andresmgsl commented

2026-07-24 12:20:59 +00:00

(Migrated from github.com)

🔎 reviewing head fa768510bf

🔎 reviewing head fa768510bff3624f40716b53c115968271015ae1

codex-bot-andresmgsl (Migrated from github.com) approved these changes 2026-07-24 12:22:14 +00:00

codex-bot-andresmgsl (Migrated from github.com) left a comment

All acceptance criteria from #134 are satisfied on fa768510bff3624f40716b53c115968271015ae1.

The implementation detects a root commit from a successful git rev-list --parents result, preserves a loud exit 128 for an unresolvable MERGE_SHA, and guards every base-scoped cat-file, fetch, and git show behind a non-empty base_sha. The tests cover all-zeros and empty EVENT_BEFORE, the parented-head fallback, the green -dev facts→decide path, the bare unlabeled refusal, and the invalid-head negative case. The changelog fragment and incident-linked comments are present.

Verified locally:

test/run.sh: 17 test files passed
.github/scripts/shellcheck-all.sh: 37 tracked scripts clean
.github/scripts/actionlint-all.sh: 6 workflow files clean
git diff --check: clean

Current GitHub checks are also green; the two release-on-event jobs are expectedly skipped for this PR.

All acceptance criteria from #134 are satisfied on `fa768510bff3624f40716b53c115968271015ae1`. The implementation detects a root commit from a successful `git rev-list --parents` result, preserves a loud exit 128 for an unresolvable `MERGE_SHA`, and guards every base-scoped `cat-file`, fetch, and `git show` behind a non-empty `base_sha`. The tests cover all-zeros and empty `EVENT_BEFORE`, the parented-head fallback, the green `-dev` facts→decide path, the bare unlabeled refusal, and the invalid-head negative case. The changelog fragment and incident-linked comments are present. Verified locally: - `test/run.sh`: 17 test files passed - `.github/scripts/shellcheck-all.sh`: 37 tracked scripts clean - `.github/scripts/actionlint-all.sh`: 6 workflow files clean - `git diff --check`: clean Current GitHub checks are also green; the two release-on-event jobs are expectedly skipped for this PR.

grok-bot-andresmgsl (Migrated from github.com) approved these changes 2026-07-24 12:22:58 +00:00

grok-bot-andresmgsl (Migrated from github.com) left a comment

Verdict: approve

Reviewed head fa768510bff3624f40716b53c115968271015ae1 against #134.

Acceptance criteria

Criterion	Result
Root commit + all-zeros `event.before` → exit 0, `base_ver=(none)`	pass — unit + manual repro
Root commit + empty `event.before` → same	pass
Root `-dev` → facts→decide green NOTICE, `ceremony=no`	pass — `release-chain.test.sh`
Root bare unlabeled → refuse exit 1, creates nothing	pass
Unresolvable `MERGE_SHA` dies loudly (128), never `(none)`	pass — D2 pin positive and negative
No `git show` / fetch with empty rev on no-base path	pass — both behind `[ -n "$base_sha" ]`; greenfield fixtures use refuse-by-default gh stub
Existing parented all-zeros / empty fallbacks still resolve to first parent	pass — prior cases unchanged; manual parented repro ok
Suite + shellcheck	pass — 17/17 test files; `shellcheck -x lib/facts.sh` clean; CI green

Spec fidelity (D1–D4)

D1: zero-parent head takes the existing (none) path; decide's table unchanged.
D2: parent count via git rev-list --parents -n 1 — fact, not || true on a failed rev-parse. Bad SHA still propagates exit 128 under set -e.
D3: belt-and-braces fetch and base git show gated on non-empty base_sha — avoids the index-read wrong-fact trap the issue names.
D4: comments carry the 0.2.0-drill / #134 war story.

Diff

Minimal and correct: lib/facts.sh, the two test files with greenfield + D2 fixtures, and changelog.d/134.md. No behavior outside the crash path changes.

Approve.

## Verdict: approve Reviewed head `fa768510bff3624f40716b53c115968271015ae1` against #134. ### Acceptance criteria | Criterion | Result | |---|---| | Root commit + all-zeros `event.before` → exit 0, `base_ver=(none)` | **pass** — unit + manual repro | | Root commit + empty `event.before` → same | **pass** | | Root `-dev` → facts→decide green NOTICE, `ceremony=no` | **pass** — `release-chain.test.sh` | | Root bare unlabeled → refuse exit 1, creates nothing | **pass** | | Unresolvable `MERGE_SHA` dies loudly (128), never `(none)` | **pass** — D2 pin positive and negative | | No `git show` / fetch with empty rev on no-base path | **pass** — both behind `[ -n "$base_sha" ]`; greenfield fixtures use refuse-by-default gh stub | | Existing parented all-zeros / empty fallbacks still resolve to first parent | **pass** — prior cases unchanged; manual parented repro ok | | Suite + shellcheck | **pass** — 17/17 test files; `shellcheck -x lib/facts.sh` clean; CI green | ### Spec fidelity (D1–D4) - **D1**: zero-parent head takes the existing `(none)` path; decide's table unchanged. - **D2**: parent count via `git rev-list --parents -n 1` — fact, not `|| true` on a failed rev-parse. Bad SHA still propagates exit 128 under `set -e`. - **D3**: belt-and-braces fetch and base `git show` gated on non-empty `base_sha` — avoids the index-read wrong-fact trap the issue names. - **D4**: comments carry the 0.2.0-drill / #134 war story. ### Diff Minimal and correct: `lib/facts.sh`, the two test files with greenfield + D2 fixtures, and `changelog.d/134.md`. No behavior outside the crash path changes. Approve.

kimi-bot-andresmgsl commented

2026-07-24 12:35:35 +00:00

(Migrated from github.com)

🔎 reviewing head fa768510bf

🔎 reviewing head fa768510bff3624f40716b53c115968271015ae1

kimi-bot-andresmgsl (Migrated from github.com) approved these changes 2026-07-24 12:36:13 +00:00

kimi-bot-andresmgsl (Migrated from github.com) left a comment

Verdict: approve

Reviewed head fa768510bff3624f40716b53c115968271015ae1 against #134, in a detached worktree.

Acceptance criteria

Criterion	Result
Root commit + all-zeros `event.before` → exit 0, `base_ver=(none)`	pass — unit test plus my own manual repro on a fresh root-commit fixture
Root commit + empty `event.before` → same	pass
Root `-dev` → `ceremony=no` green NOTICE through facts→decide	pass — `release-chain.test.sh` greenfield cases, 8/8
Root bare `0.1.0` unlabeled → refuse exit 1, creates nothing	pass — e2e with refuse-by-default gh stub
Unresolvable `MERGE_SHA` dies loudly, never reported as `(none)`	pass — D2 pin asserts both exit 128 "bad object" and the absence of `base_ver=(none)`; reproduced manually (grep count 0)
No `git show` / `cat-file` / `fetch` with an empty rev on the no-base path	pass — belt-and-braces fetch and base `git show` both behind `[ -n "$base_sha" ]`; the index-read hazard from D3 is commented
Existing facts tests pass unchanged	pass — 24 pass; the only 2 failures are `package-json:` cases needing `node`, absent in this environment and green in CI
Suite + shellcheck + actionlint	pass — same node-only local exceptions; CI fully green at this head (test, self-guards, release-exercise, action-exercise, docs-sync-exercise)

Spec fidelity (D1–D4)

D1: a parentless head takes the existing base_ver=(none) path; decide's table is untouched — a -dev root is the green bootstrap, a bare root still refuses.
D2: parent count read as a fact via git rev-list --parents -n 1; under set -e an unresolvable head still propagates the 128 — detection, not inference from a failed command. The *" "* split is the right discriminator.
D3: the no-base path runs no rev-scoped command with an empty rev; the comment names the index-read trap (git show ":VERSION"), which is the reason this invariant matters.
D4: comments carry the 0.2.0-drill war story and #134, per CONTRIBUTING.

Conventions

One fragment changelog.d/134.md, published prose only; no CHANGELOG.md edit.
The head's retrigger commit is empty; the change itself is two commits (fix + tests).

Notes (non-blocking)

None blocking. The diff is exactly the scoped fix plus its tests — no drive-by.

Meets every #134 criterion. Approve.

## Verdict: approve Reviewed head `fa768510bff3624f40716b53c115968271015ae1` against #134, in a detached worktree. ### Acceptance criteria | Criterion | Result | |---|---| | Root commit + all-zeros `event.before` → exit 0, `base_ver=(none)` | **pass** — unit test plus my own manual repro on a fresh root-commit fixture | | Root commit + empty `event.before` → same | **pass** | | Root `-dev` → `ceremony=no` green NOTICE through facts→decide | **pass** — `release-chain.test.sh` greenfield cases, 8/8 | | Root bare `0.1.0` unlabeled → refuse exit 1, creates nothing | **pass** — e2e with refuse-by-default gh stub | | Unresolvable `MERGE_SHA` dies loudly, never reported as `(none)` | **pass** — D2 pin asserts both exit 128 "bad object" and the absence of `base_ver=(none)`; reproduced manually (grep count 0) | | No `git show` / `cat-file` / `fetch` with an empty rev on the no-base path | **pass** — belt-and-braces fetch and base `git show` both behind `[ -n "$base_sha" ]`; the index-read hazard from D3 is commented | | Existing facts tests pass unchanged | **pass** — 24 pass; the only 2 failures are `package-json:` cases needing `node`, absent in this environment and green in CI | | Suite + shellcheck + actionlint | **pass** — same node-only local exceptions; CI fully green at this head (test, self-guards, release-exercise, action-exercise, docs-sync-exercise) | ### Spec fidelity (D1–D4) - **D1**: a parentless head takes the existing `base_ver=(none)` path; `decide`'s table is untouched — a `-dev` root is the green bootstrap, a bare root still refuses. - **D2**: parent count read as a fact via `git rev-list --parents -n 1`; under `set -e` an unresolvable head still propagates the 128 — detection, not inference from a failed command. The `*" "*` split is the right discriminator. - **D3**: the no-base path runs no rev-scoped command with an empty rev; the comment names the index-read trap (`git show ":VERSION"`), which is the reason this invariant matters. - **D4**: comments carry the 0.2.0-drill war story and #134, per CONTRIBUTING. ### Conventions - One fragment `changelog.d/134.md`, published prose only; no `CHANGELOG.md` edit. - The head's retrigger commit is empty; the change itself is two commits (fix + tests). ### Notes (non-blocking) - None blocking. The diff is exactly the scoped fix plus its tests — no drive-by. Meets every #134 criterion. Approve.

claude-bot-andresmgsl commented

2026-07-24 12:37:49 +00:00

(Migrated from github.com)

Handoff — round converged, over to @danmt

What shipped. lib/facts.sh no longer dies at exit 128 when the merge head is a root commit — the shape both 0.2.0 drills hit on a repository's first push to main (event.before all-zeros, MERGE_SHA^1 nonexistent). Per #134's decided spec: the head's parent count is read as a fact via git rev-list --parents -n 1 (D2); a zero-parent head takes the existing base_ver="(none)" path (D1) with the belt-and-braces fetch and the base git show skipped, so no rev-scoped command ever runs with an empty rev (D3); the comments carry the 0.2.0-drill war story (D4). An unresolvable MERGE_SHA still fails loudly at 128 — it never masquerades as (none).

Rounds. One. The panel (codex, grok, kimi) was requested at 12:16Z on head fa76851; all three approved that head with no changes requested, so the round converged whole on the first pass — no post-review fix commits, and the reviewed head is the merge head.

What was verified.

All 8 acceptance criteria from #134, checked in the PR body and independently confirmed by each reviewer; grok and kimi additionally reproduced the greenfield fixtures by hand (kimi in a detached worktree).
New coverage: test/facts.test.sh greenfield fixtures (all-zeros and empty event.before, bare root with labeled-no) plus the D2 loud-death pin (bad MERGE_SHA → 128, no (none) in output); test/release-chain.test.sh end-to-end greenfield cases (root -dev goes green ceremony=no with a NOTICE; bare unlabeled root still refuses exit 1). All pre-existing facts.test.sh cases pass unchanged.
CI fully green on fa76851: test suite (17 files), shellcheck sweep (37 scripts), actionlint, labels/scope, labels/reconcile, release-exercise (fixture-chain + both step-replays), self-guards, action-exercise, docs-sync-exercise.

Residue / notes.

The brief blocker:ci-red at 12:17Z was false — a queued duplicate labels/reconcile run was cancelled and read as red (same shape as #138); the empty-commit retrigger fa76851 cleared it and became the reviewed head. No underlying red ever existed.
No post-merge steps owed. The change is self-contained; changelog.d/134.md publishes with the next release PR's assembly.
The reconciler validated the convergence at 12:36Z and applied state:needs-human + the review request itself, so this summary completes the handoff.

This PR could be merged right now. The merge is yours.

## Handoff — round converged, over to @danmt **What shipped.** `lib/facts.sh` no longer dies at exit 128 when the merge head is a root commit — the shape both 0.2.0 drills hit on a repository's first push to `main` (`event.before` all-zeros, `MERGE_SHA^1` nonexistent). Per #134's decided spec: the head's parent count is read as a fact via `git rev-list --parents -n 1` (D2); a zero-parent head takes the existing `base_ver="(none)"` path (D1) with the belt-and-braces fetch and the base `git show` skipped, so no rev-scoped command ever runs with an empty rev (D3); the comments carry the 0.2.0-drill war story (D4). An unresolvable `MERGE_SHA` still fails loudly at 128 — it never masquerades as `(none)`. **Rounds.** One. The panel (codex, grok, kimi) was requested at 12:16Z on head `fa76851`; all three approved that head with no changes requested, so the round converged whole on the first pass — no post-review fix commits, and the reviewed head is the merge head. **What was verified.** - All 8 acceptance criteria from #134, checked in the PR body and independently confirmed by each reviewer; grok and kimi additionally reproduced the greenfield fixtures by hand (kimi in a detached worktree). - New coverage: `test/facts.test.sh` greenfield fixtures (all-zeros and empty `event.before`, bare root with `labeled-no`) plus the D2 loud-death pin (bad `MERGE_SHA` → 128, no `(none)` in output); `test/release-chain.test.sh` end-to-end greenfield cases (root `-dev` goes green `ceremony=no` with a NOTICE; bare unlabeled root still refuses exit 1). All pre-existing `facts.test.sh` cases pass unchanged. - CI fully green on `fa76851`: test suite (17 files), shellcheck sweep (37 scripts), actionlint, labels/scope, labels/reconcile, release-exercise (fixture-chain + both step-replays), self-guards, action-exercise, docs-sync-exercise. **Residue / notes.** - The brief `blocker:ci-red` at 12:17Z was false — a queued duplicate `labels/reconcile` run was cancelled and read as red (same shape as #138); the empty-commit retrigger `fa76851` cleared it and became the reviewed head. No underlying red ever existed. - No post-merge steps owed. The change is self-contained; `changelog.d/134.md` publishes with the next release PR's assembly. - The reconciler validated the convergence at 12:36Z and applied `state:needs-human` + the review request itself, so this summary completes the handoff. This PR could be merged right now. The merge is yours.