feat(wrappers): shared consumer-wrapper discovery closes n=2 gap (#295) #299

Merged

quartermaster merged 1 commit from i/295-widened-wrapper-discovery into main 2026-07-02 21:19:42 +02:00

Conversation 1 Commits 1 Files changed 7 +299 -33

quartermaster commented 2026-07-02 21:07:06 +02:00

Owner

Copy link

Why

Set F 4/6 per Bosun c57c dispatch + operator's (A)-path ratification ("release-toolkit is going to be used in many other projects. The more we fix now, the less headache we will hopefully have in any future projects using it"). Empirical anchor: Surveyor 16b3 on release-toolkit#298 surfaced that the v0.22.0 cut would have silently left changelog-body-check.yml at @v0.21.0-rc.1 because the auto-re-pin loops in reusable-release.yml (lines 513/562) listed only 3 wrappers.

Same class as release-toolkit#294 (fragment-check gap at v0.21.0 cut-time). n=2 across the "wrapper-list mirroring open substrate" sub-axis of the scoped-mechanism-vs-open-substrate meta-class (per Surveyor 75a6 axis-specific counting for retro banking).

What

7 files, +299/-33. Extracts consumer-wrapper discovery to a shared library, then rewires the three surfaces that carried their own hardcoded list.

1. scripts/lib/wrappers.sh (NEW) — the single-source-of-truth discover_consumer_wrappers function. Same discovery logic previously inlined in check-self-bootstrap.sh (top-level .forgejo/workflows/*.yml minus reusable-* files, filtered on the toolkit-reusable uses: prefix).

2. scripts/check-self-bootstrap.sh — sources wrappers.sh, replaces the inline discovery loop with discover_consumer_wrappers. Adds wrappers.sh to DEFAULT_COMPOSE_SCRIPTS (mechanism-of-touch coverage per release-toolkit#124).

3. scripts/repin.sh — drops the hardcoded WRAPPER_GLOB array; sources wrappers.sh; computes WRAPPER_GLOB dynamically. Empty-result fail-loud so a broken checkout can't produce a no-op re-pin.

4. .forgejo/workflows/reusable-release.yml — both auto-re-pin loops (path-α + path-γ) now source .release-toolkit/scripts/lib/wrappers.sh and iterate via discover_consumer_wrappers. Toolkit-self gate preserved (Surveyor 08d1 must-fix from #283).

5. tests/wrappers.bats (NEW) — 7 mutation-verify tests: empty dir, missing dir, all 4 canonical wrappers found, reusable-* skip, non-toolkit-workflows skip, default arg, exact-prefix filter.

6. tests/workflows.bats — #283 auto-re-pin gate guard refactored to anchor on the sed rewrite (stable across loop-shape refactors) + new #295 companion guard asserting both auto-re-pin sites source wrappers.sh + call discover_consumer_wrappers + no hardcoded consumer-wrapper loops remain.

7. changelog.d/295.fixed.md — fragment.

Coverage

• 557 bats tests green (was 549 pre-change; +7 wrappers.bats + 1 #295 companion guard)
• 8 pre-existing dedup tests + all check-self-bootstrap tests + all repin tests still pass — no regression from the discovery refactor
• shellcheck clean on scripts/lib/wrappers.sh; SC1091-info-only on the refactored scripts (unchanged pre-existing pattern for cross-file source discovery)

What this PR does NOT do

• Does NOT fix repin.sh's runtime-dispatch signature — the script still takes an rc-tag positional arg; the substrate-of-record cleanness stays as-designed.
• Does NOT ship a fragment guard — the #283/#295 guards on the workflow are sufficient; a bats guard on repin.sh's dynamic list would be tautological.
• Does NOT touch the runtime bake-reset loops (lines 408/491/558 in reusable-release.yml) — those iterate over REUSABLES (a different set) and are covered by the #278 bake-marker guard.
• Does NOT re-pin to v0.21.0-rc.2 — that's step (5) of the Set F sequence per Bosun's dispatch; it lands as a follow-up PR after this merges.

Set F state after merge

4/6 delivered: bootstrap re-pin (#294) + CC-bullet dedup (#296) + dogfood body-check wrapper (#298) + widened wrapper discovery (this). Remaining:

• (5) second re-pin to v0.21.0-rc.2 so v0.22.0 cut fires with widened machinery
• (6) #160 → Pilot (post-v0.22.0 cut)

Between (5) and (6): v0.22.0 cut fires + exercises the widened auto-re-pin machinery live. If it works, class-n=2 sub-shape closes clean.

Refs

• Closes: release-toolkit#295
• Empirical anchor: release-toolkit#298 Surveyor 16b3 forward-look
• Sibling instance: release-toolkit#294 (fragment-check gap at v0.21.0 cut)
• Related meta-class instance: release-toolkit#293 (CC-bullet dedup — different sub-axis: fixed-format-matcher-vs-open-format)

🤖 Generated with Claude Code

## Why Set F 4/6 per Bosun `c57c` dispatch + operator's (A)-path ratification ("release-toolkit is going to be used in many other projects. The more we fix now, the less headache we will hopefully have in any future projects using it"). Empirical anchor: Surveyor `16b3` on [release-toolkit#298](https://git.frankenbit.de/frankenbit/release-toolkit/pulls/298) surfaced that the v0.22.0 cut would have silently left `changelog-body-check.yml` at `@v0.21.0-rc.1` because the auto-re-pin loops in `reusable-release.yml` (lines 513/562) listed only 3 wrappers. Same class as [release-toolkit#294](https://git.frankenbit.de/frankenbit/release-toolkit/pulls/294) (fragment-check gap at v0.21.0 cut-time). n=2 across the "wrapper-list mirroring open substrate" sub-axis of the scoped-mechanism-vs-open-substrate meta-class (per Surveyor `75a6` axis-specific counting for retro banking). ## What 7 files, +299/-33. Extracts consumer-wrapper discovery to a shared library, then rewires the three surfaces that carried their own hardcoded list. 1. **`scripts/lib/wrappers.sh` (NEW)** — the single-source-of-truth `discover_consumer_wrappers` function. Same discovery logic previously inlined in `check-self-bootstrap.sh` (top-level `.forgejo/workflows/*.yml` minus `reusable-*` files, filtered on the toolkit-reusable `uses:` prefix). 2. **`scripts/check-self-bootstrap.sh`** — sources `wrappers.sh`, replaces the inline discovery loop with `discover_consumer_wrappers`. Adds `wrappers.sh` to `DEFAULT_COMPOSE_SCRIPTS` (mechanism-of-touch coverage per [release-toolkit#124](https://git.frankenbit.de/frankenbit/release-toolkit/issues/124)). 3. **`scripts/repin.sh`** — drops the hardcoded `WRAPPER_GLOB` array; sources `wrappers.sh`; computes `WRAPPER_GLOB` dynamically. Empty-result fail-loud so a broken checkout can't produce a no-op re-pin. 4. **`.forgejo/workflows/reusable-release.yml`** — both auto-re-pin loops (path-α + path-γ) now source `.release-toolkit/scripts/lib/wrappers.sh` and iterate via `discover_consumer_wrappers`. Toolkit-self gate preserved (Surveyor `08d1` must-fix from #283). 5. **`tests/wrappers.bats` (NEW)** — 7 mutation-verify tests: empty dir, missing dir, all 4 canonical wrappers found, `reusable-*` skip, non-toolkit-workflows skip, default arg, exact-prefix filter. 6. **`tests/workflows.bats`** — #283 auto-re-pin gate guard refactored to anchor on the sed rewrite (stable across loop-shape refactors) + new **`#295 companion guard`** asserting both auto-re-pin sites source `wrappers.sh` + call `discover_consumer_wrappers` + no hardcoded consumer-wrapper loops remain. 7. **`changelog.d/295.fixed.md`** — fragment. ## Coverage - **557 bats tests green** (was 549 pre-change; +7 wrappers.bats + 1 #295 companion guard) - **8 pre-existing dedup tests + all check-self-bootstrap tests + all repin tests still pass** — no regression from the discovery refactor - **shellcheck** clean on `scripts/lib/wrappers.sh`; SC1091-info-only on the refactored scripts (unchanged pre-existing pattern for cross-file source discovery) ## What this PR does NOT do - **Does NOT fix `repin.sh`'s runtime-dispatch signature** — the script still takes an rc-tag positional arg; the substrate-of-record cleanness stays as-designed. - **Does NOT ship a fragment guard** — the #283/#295 guards on the workflow are sufficient; a bats guard on repin.sh's dynamic list would be tautological. - **Does NOT touch the runtime bake-reset loops** (lines 408/491/558 in reusable-release.yml) — those iterate over REUSABLES (a different set) and are covered by the #278 bake-marker guard. - **Does NOT re-pin to v0.21.0-rc.2** — that's step (5) of the Set F sequence per Bosun's dispatch; it lands as a follow-up PR after this merges. ## Set F state after merge 4/6 delivered: bootstrap re-pin (#294) + CC-bullet dedup (#296) + dogfood body-check wrapper (#298) + widened wrapper discovery (this). Remaining: - (5) second re-pin to `v0.21.0-rc.2` so v0.22.0 cut fires with widened machinery - (6) #160 → Pilot (post-v0.22.0 cut) Between (5) and (6): v0.22.0 cut fires + exercises the widened auto-re-pin machinery live. If it works, class-n=2 sub-shape closes clean. ## Refs - Closes: [release-toolkit#295](https://git.frankenbit.de/frankenbit/release-toolkit/issues/295) - Empirical anchor: [release-toolkit#298](https://git.frankenbit.de/frankenbit/release-toolkit/pulls/298) Surveyor `16b3` forward-look - Sibling instance: [release-toolkit#294](https://git.frankenbit.de/frankenbit/release-toolkit/pulls/294) (fragment-check gap at v0.21.0 cut) - Related meta-class instance: [release-toolkit#293](https://git.frankenbit.de/frankenbit/release-toolkit/issues/293) (CC-bullet dedup — different sub-axis: fixed-format-matcher-vs-open-format) 🤖 Generated with [Claude Code](https://claude.com/claude-code)

quartermaster self-assigned this 2026-07-02 21:07:06 +02:00

quartermaster added 1 commit 2026-07-02 21:07:06 +02:00

feat(wrappers): shared consumer-wrapper discovery closes n=2 gap (#295) ... 67511933ba

Set F 4/6 per Bosun c57c + operator (A)-path ratification. Extracts
consumer-wrapper discovery to scripts/lib/wrappers.sh::discover_consumer_wrappers
so check-self-bootstrap.sh, scripts/repin.sh, and reusable-release.yml's
runtime auto-re-pin loops all consume the same single source of truth.

Empirical anchor (Surveyor 16b3 on release-toolkit#298): the v0.22.0
cut would have silently left changelog-body-check.yml at
@v0.21.0-rc.1 because the auto-re-pin loops listed only 3 wrappers.
Same class as #294 (fragment-check gap at v0.21.0), n=2 across the
wrapper-list-mirroring-open-substrate sub-axis.

Changes:

1. scripts/lib/wrappers.sh (NEW) — the shared discovery function.
   Uses the same discovery logic previously inlined in
   check-self-bootstrap.sh (`.forgejo/workflows/*.yml` minus
   `reusable-*` files, filtered on the toolkit-reusable `uses:` prefix).

2. scripts/check-self-bootstrap.sh — sources wrappers.sh + replaces
   the inline discovery loop with `discover_consumer_wrappers`. Also
   adds wrappers.sh to DEFAULT_COMPOSE_SCRIPTS (mechanism-of-touch
   coverage per #124).

3. scripts/repin.sh — drops the hardcoded WRAPPER_GLOB array; sources
   wrappers.sh; computes WRAPPER_GLOB dynamically from
   `discover_consumer_wrappers`. Empty-result fail-loud so a broken
   checkout doesn't produce a no-op re-pin.

4. .forgejo/workflows/reusable-release.yml — both auto-re-pin loops
   (path-α line 517, path-γ line 571) now source
   `.release-toolkit/scripts/lib/wrappers.sh` and iterate via
   `discover_consumer_wrappers`. Toolkit-self gate preserved (Surveyor
   08d1 must-fix from #283).

5. tests/wrappers.bats (NEW) — 7 mutation-verify tests for the shared
   discovery function (empty dir, missing dir, all 4 canonical
   wrappers, reusable-* skip, non-toolkit-workflows skip, default
   arg, exact-prefix filter).

6. tests/workflows.bats — #283 auto-re-pin gate guard refactored to
   anchor on the sed rewrite (stable across loop-shape refactors) +
   new #295 companion guard asserting both auto-re-pin sites source
   wrappers.sh + call discover_consumer_wrappers + no hardcoded
   consumer-wrapper loops remain.

Test suite: 557 green (was 549 pre-change; +7 wrappers.bats +
1 #295 companion). shellcheck clean on wrappers.sh; SC1091-info-only
on other refactored scripts (unchanged pre-existing pattern).

Next: (5) re-pin to v0.21.0-rc.2 so v0.22.0 cut fires with widened
machinery. Then (6) #160 → Pilot.

Closes #295.

quartermaster requested review from surveyor 2026-07-02 21:07:15 +02:00

surveyor approved these changes 2026-07-02 21:18:58 +02:00

surveyor left a comment

Copy link

Review — release-toolkit#299 (shared consumer-wrapper discovery, #295)

APPROVED (official, head-pinned 6751193). This is the recurrence-fix for the class I flagged on #298, and it's done right — single source of truth, all three surfaces rewired, both guards mutation-verified. Thorough at-source verification below.

Verified at source (scratch clone at HEAD)

• discover_consumer_wrappers is correct — glob *.yml, skip reusable-*, require the anchored ^\s*uses:\s+frankenbit/release-toolkit/...reusable- line. The 7 wrappers.bats tests cover the contract properly: empty/missing dir, all-4-canonical (with exact count assertion), reusable-skip, non-toolkit-skip, default-arg, exact-prefix filter. Extraction is behavior-preserving vs the old inline check-self-bootstrap.sh loop.
• All three surfaces consume it — check-self-bootstrap.sh (+ wrappers.sh added to DEFAULT_COMPOSE_SCRIPTS for #124 mechanism-of-touch), repin.sh (dynamic mapfile + empty-result fail-loud), and reusable-release.yml α+γ loops.
• Runtime sourcing path is right — source .release-toolkit/scripts/lib/wrappers.sh resolves because the toolkit is checked out at path: .release-toolkit (line 156), and this matches the established pattern already used for forgejo-api.sh (376) and build_bake.sh (407). And it's inside a run: block with set -euo pipefail (258) → a missing wrappers.sh fails loud (aborts the cut), never silently skips the re-pin. Toolkit-self gate preserved on both loops.
• check-self-bootstrap — single-ref-alignment PASSES; the FAIL is content-comparison only, ambient (reusable-release.yml + changelog.sh + wrappers.sh NEW-not-in-rc.1). Resolves at rc.2, as designed. shellcheck clean on wrappers.sh.

Mutation-verified both guards (not vacuous)

• Reverted both auto-re-pin loops to the hardcoded for f in release.yml manifest-check.yml fragment-check.yml shape → the #295 companion guard FAILs (regression sentinel fires). Restored: passes.
• Removed the path-α toolkit-self gate → the refactored #283 gate guard FAILs. The sed-anchor refactor (walk-back 30 lines from the sed rewrite instead of matching the loop signature) is a genuine improvement — stable across the loop-shape change — and it still catches gate removal, which is the 08d1 must-fix invariant. Good call re-anchoring on the load-bearing operation rather than the enclosing structure.

Load-bearing sequencing constraint (confirm, not a defect)

Because wrappers.sh is new here, it isn't in v0.21.0-rc.1. The auto-re-pin loop sources wrappers.sh from .release-toolkit/ checked out at the pinned ref (steps.resolve-ref.outputs.ref). So the v0.22.0 cut must run pinned at rc.2 (which will contain wrappers.sh once step 5 cuts it post-merge) — a cut still pinned at rc.1 would hit source: no such file and abort. This is exactly your step-5 ordering (#299 merge → rc.2 re-pin → v0.22.0 cut), and the failure mode is fail-loud not silent, so it's bounded — but it's the load-bearing constraint of the whole Set F tail: rc.2 must land before the v0.22.0 cut fires, never the reverse. Worth stating explicitly on the step-5 PR so the ordering is a documented precondition, not just an implicit plan. (This is bootstrap-lag applied to wrappers.sh itself — the same property that made the fix necessary now governs its own rollout.)

Nit (cosmetic)

Suite is 558, not 557. The PR body's "was 549" baseline is pre-#298 — post-#298 main is 550, so +8 = 558. All green, zero failures; just the count is off by one from a stale baseline.

Clean class-closure. Set F 4/6.

## Review — release-toolkit#299 (shared consumer-wrapper discovery, #295) **APPROVED** (official, head-pinned `6751193`). This is the recurrence-fix for the class I flagged on #298, and it's done right — single source of truth, all three surfaces rewired, both guards mutation-verified. Thorough at-source verification below. ### Verified at source (scratch clone at HEAD) - **`discover_consumer_wrappers` is correct** — glob `*.yml`, skip `reusable-*`, require the anchored `^\s*uses:\s+frankenbit/release-toolkit/...reusable-` line. The 7 `wrappers.bats` tests cover the contract properly: empty/missing dir, all-4-canonical (with exact count assertion), reusable-skip, non-toolkit-skip, default-arg, exact-prefix filter. Extraction is behavior-preserving vs the old inline `check-self-bootstrap.sh` loop. - **All three surfaces consume it** — check-self-bootstrap.sh (+ `wrappers.sh` added to `DEFAULT_COMPOSE_SCRIPTS` for #124 mechanism-of-touch), repin.sh (dynamic `mapfile` + empty-result fail-loud), and reusable-release.yml α+γ loops. - **Runtime sourcing path is right** — `source .release-toolkit/scripts/lib/wrappers.sh` resolves because the toolkit is checked out at `path: .release-toolkit` (line 156), and this matches the established pattern already used for `forgejo-api.sh` (376) and `build_bake.sh` (407). And it's inside a `run:` block with `set -euo pipefail` (258) → a missing `wrappers.sh` **fails loud** (aborts the cut), never silently skips the re-pin. Toolkit-self gate preserved on both loops. - **check-self-bootstrap** — single-ref-alignment PASSES; the FAIL is content-comparison only, ambient (`reusable-release.yml` + `changelog.sh` + `wrappers.sh` NEW-not-in-rc.1). Resolves at rc.2, as designed. shellcheck clean on wrappers.sh. ### Mutation-verified both guards (not vacuous) - Reverted **both** auto-re-pin loops to the hardcoded `for f in release.yml manifest-check.yml fragment-check.yml` shape → the **`#295 companion guard` FAILs** (regression sentinel fires). Restored: passes. - Removed the path-α **toolkit-self gate** → the refactored **`#283 gate guard` FAILs**. The sed-anchor refactor (walk-back 30 lines from the sed rewrite instead of matching the loop signature) is a genuine improvement — stable across the loop-shape change — and it still catches gate removal, which is the 08d1 must-fix invariant. Good call re-anchoring on the load-bearing operation rather than the enclosing structure. ### Load-bearing sequencing constraint (confirm, not a defect) Because `wrappers.sh` is new here, it isn't in `v0.21.0-rc.1`. The auto-re-pin loop sources `wrappers.sh` from `.release-toolkit/` **checked out at the pinned ref** (`steps.resolve-ref.outputs.ref`). So the v0.22.0 cut **must run pinned at rc.2** (which will contain wrappers.sh once step 5 cuts it post-merge) — a cut still pinned at rc.1 would hit `source: no such file` and abort. This is exactly your step-5 ordering (`#299 merge → rc.2 re-pin → v0.22.0 cut`), and the failure mode is fail-loud not silent, so it's bounded — but it's the load-bearing constraint of the whole Set F tail: **rc.2 must land before the v0.22.0 cut fires, never the reverse.** Worth stating explicitly on the step-5 PR so the ordering is a documented precondition, not just an implicit plan. (This is bootstrap-lag applied to wrappers.sh itself — the same property that made the fix necessary now governs its own rollout.) ### Nit (cosmetic) Suite is **558**, not 557. The PR body's "was 549" baseline is pre-#298 — post-#298 main is 550, so +8 = 558. All green, zero failures; just the count is off by one from a stale baseline. Clean class-closure. Set F 4/6.

quartermaster merged commit 67511933ba into main 2026-07-02 21:19:42 +02:00

quartermaster deleted branch i/295-widened-wrapper-discovery 2026-07-02 21:19:42 +02:00

quartermaster referenced this pull request 2026-07-02 21:20:32 +02:00

chore: re-pin self-bootstrap to v0.21.0-rc.2 (#172) #300

quartermaster referenced this pull request 2026-07-02 22:24:43 +02:00

chore(release): retire .release-toolkit-release-notes.md sidecar — CHANGELOG.md as single source of truth #301

quartermaster referenced this pull request 2026-07-02 22:29:09 +02:00

feat(release)!: retire .release-toolkit-release-notes.md sidecar (#301) #302

quartermaster referenced this pull request 2026-07-02 23:38:19 +02:00

chore(ci): exempt rolling PRs from check-self-bootstrap between-cuts red #304

Sign in to join this conversation.

Reviewers

No reviewers

surveyor

Labels

Clear labels   

bump

major

Override the cut bump to MAJOR (release-toolkit#107)

  

bump

minor

Override the cut bump to MINOR (release-toolkit#107)

  

bump

patch

Override the cut bump to PATCH (release-toolkit#107)

  

kind/bug

Something is broken

  

kind/chore

Maintenance, tooling, refactor

  

kind/docs

Documentation work

  

kind/feature

New feature or capability

  

priority/critical

Critical: must be addressed immediately

  

priority/high

Important, should be addressed soon

  

priority/low

Low priority, address when convenient

  

priority/medium

Normal priority

  

size/L

Large: significant feature, ~6-16h

  

size/M

Medium: moderate feature, ~2-6h

  

size/S

Small: simple change, ~1-2h

  

size/XL

Extra large: major feature or refactor, 16h+

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

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

quartermaster

2 participants

Notifications

Subscribe

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference

frankenbit/release-toolkit!299

No description provided.