test(bundle): drive tide's verdict arms (unknown-retry exhaustion, moved head, 405/409 refusals, merged/closed/locked, check_suite guards, once-per-run) through dist/index.js

[hivecommons-hive]

Test Improvement

Adds __tests__/bundle/tideVerdictArms.test.ts (13 tests, new file only — no production, harness or config change), driving src/plugins/tide.ts's verdict arms through dist/index.js against the fake GitHub:

• fetchMergeability: mergeability still unknown after the 1 s / 2 s / 4 s retries → skipping pr #1: not mergeable (unknown) (lines 99, 108–111)
• a head that moves while mergeability is computed → head moved during evaluation (208)
• a merge refused with 405 once a concurrent event merged it → pr #1 was merged concurrently, exit 0 (219)
• a merge refused with 409 → base branch moved / head moved (223, both regex arms)
• a refused merge whose re-read fails too → could not merge pr #1: …, run fails (439)
• blockedReason: a pull request GitHub reports as already merged, closed, locked, skipped before its binding or mergeability is read (399, 402, 405)
• tideOnComment on a closed pull request after /hold (496–497)
• tideOnCheckSuite without a head sha (520) and naming no pull request with the listing finding none (564–565)
• evaluateMerge once: schedule with jobs: 'sweep lgtm' merges the shared pull request once and the other job skips it as already evaluated in this run (175)

check_suite with pull_requests: [{ number: 1 }] is the vehicle for the merge-path arms: tide is its only handler, so every recorded request is tide's and the sequences are asserted exactly (including the one POST /graphql dequeue check a gate skip makes on an event, since tide.merge_queue defaults to auto).

Measured with npm run test:coverage:e2e on main @ 3fc21f2: src/plugins/tide.ts 158/195 → 173/195 lines from __tests__/bundle alone. npx vitest run __tests__/bundle/tideVerdictArms.test.ts passes (13/13); npx eslint clean; npm run test:coverage passes all 1652 tests (the only red is the pre-existing unhandled ERR_TLS_CERT_ALTNAME_INVALID from __tests__/issueCommentTest/meow.test.ts, tracked in #294 and fixed by #287).

Disjoint from every open __tests__/bundle PR: these lines stay uncovered with all 33 held bundle PRs (#242–#345) applied together, and the file name is new. It does not use #246's repoLabels helper so it does not depend on that PR landing.

Related Issue

Closes #346

---

Filed by quality agent (hold-gated mode). Human review required.

— hive: agent=quality backend=copilot model=claude-fable-5.1

[hivecommons-hive]

Important

Held for human review by the hive's ACMM level gate.

This PR was opened by the "quality" agent while Hive policy required a human checkpoint for that agent. Non-outreach agents are held at ACMM L3–L5; the outreach agent is always held because it publishes project-facing communication.

Hive will keep the hold label until a human removes it. Operators can make a deliberate one-off release during an ACMM level change with release_level_holds=true, but level changes never release this hold automatically.

[jeefy]

All 13 cases add coverage.

• tideVerdictArms.test.ts:159–162: use toHaveLength(2) plus fragments (merge exploded, could not merge pull request(s) #1) instead of comparing both ::error:: lines with toEqual.
• Not blocking: the L77 case spends about 7.3s waiting through real 1s+2s+4s retries, which makes this the slowest bundle file. If there's a way to shorten the backoff for tests, use it.

[hivecommons-hive]

Addressed: the two ::error:: lines are now toHaveLength(2) plus the merge exploded / could not merge pull request(s) #1 fragments. On the backoff: the delays are hard-coded in tide.ts with no test hook, and this PR only touches the test file, so I've left the ~7s case as is rather than reach into src/. Merged main too (it brings #287's meow network fix, which is what build-test was failing on).

---

🐝 Hive Agent: quality | Instance: hosted-available-lke648397-260827-5q9t | SHA: unknown

— hive: agent=quality backend=copilot model=claude-fable-5.1

[hivecommons-hive]

Addressed in 3acb2a0: toHaveLength(2) plus merge exploded / could not merge pull request(s) #1 fragments. On the 7.3s backoff: unknownRetryDelaysMs is a const in src/plugins/tide.ts with no test-side override, so shortening it needs a production change (e.g. an env or input knob) that this PR deliberately doesn't make; happy to follow up if you want one.

CI is green on the new head.

---

🐝 Hive Agent: quality | Instance: hosted-available-lke648397-260827-5q9t | SHA: 1f7d33c

— hive: agent=quality backend=copilot model=claude-fable-5.1 copilot=1.0.88

[jeefy]

The error lines are now fragments. On current main one row adds nothing:

• tideVerdictArms.test.ts:61: drop the ['closed', …] row, and remove "closed" from the header comment at :14. #363 merged after the first review, and approveReviewShortCircuits.test.ts:72 now reaches tide.ts:401–402.
• Optional: the 7s backoff can be shortened without touching src. meow.test.ts:78 already loads a test-only script into the child process with NODE_OPTIONS=--require, and a similar script could shorten the 1000/2000/4000ms setTimeout waits.

[hivecommons-hive]

Addressed in 5235110:

• dropped the closed row (approveReviewShortCircuits.test.ts:72 reaches tide.ts:401–402 since test(bundle): drive syncApprovalReview's not-open and nothing-to-dismiss short-circuits and withdrawalReason's no-files and lgtm-skip arms through dist/index.js #363) and removed "closed" from the header comment.
• took the optional suggestion: shortBackoffPreload.cjs is loaded into the check_suite child with NODE_OPTIONS=--require (same shape as catApiPreload.cjs) and maps the setTimeout delays listed in SHORT_BACKOFF_MS=1000,2000,4000 to 10ms; every other timer is untouched and the logged retrying in Nms lines are unchanged. The file now runs in ~3.6s (the L77 case drops from ~7.3s to ~0.4s).

---

🐝 Hive Agent: quality | Instance: hosted-available-lke648397-260827-5q9t | SHA: 5235110

— hive: agent=quality backend=copilot model=claude-fable-5.1 copilot=1.0.88

[mrbobbytables]

/kind cleanup