Repository navigation
fix(ci): clear two main-red pins left by same-day merges - #14520
Pearcekieser wants to merge 1 commit into
Conversation
kirodotdev#14097 added the bn metrics_card_old_data value with Bengali digits, taking the bn catalog to 9 against the numerals ratchet of 8; the value now uses Western digits, as style/bn.md section 2 asks. kirodotdev#13883 added a list-every-site pin over gateway-supervisor.js that predates the restart successor-ownership check and the boot refuse-to-adopt check from kirodotdev#13344, so both fail it as unclaimed. Both are identity judgements (is the holder this app, or someone else), so they join IDENTITY_SITES. No runtime code changes.
UX Review (Fable 5, fork) — ✅ PASSUX-level review of UX-Verdict: PASS Nothing to check. [UX-REVIEWED] 78ad3d7 |
GPT 5.6 Review (fork) — ✅ no blocking findingsReviewed Review detailsNo findings. |
First Principles Review (Fable 5, fork) — ✅ PASSPremise-level review of First-Principles-Verdict: PASS Verify the PR's own Electron CI is green — the fail-closed pin will itself surface any #13344 site the six entries still miss. What this change shipsInventory (2 items) — 2 justifiedIntent: unwedge main's two red CI pins so open PRs can reach readiness again — a FIX.
Evidence checked against base: [FIRST-PRINCIPLES-REVIEWED] 78ad3d7 |
Design Review (Fable 5, fork) —
|
Opus 5 Review (fork) — ✅ no blocking findingsReviewed |
|
Closing as superseded by #14516, which fixes the same two main-red pins and was opened first in this repo. |
Problem / Motivation
Goal: main's Frontend Tests and Electron Shell Tests pass again, so they stop failing readiness on every open PR.
Since the merges at 17:18Z on 2026-09-27, main CI (run 36337451146 on
20ed6a1dc7) fails two pins. Frontend Tests shard 2 fails the bn numerals ratchet with 9 violations against a limit of 8. Electron Shell Tests fails two list-every-site checks inport-owner-self-asserted-identity.test.js.Why it matters
Every open PR inherits both failures, so none of them can reach a green readiness verdict until main is fixed.
Not a goal
gateway-supervisor.js, or reclassifying how it decides port ownership.What changed (motivation → approach → change)
The bn failure came from #14097, which added
app.metrics_card_old_datawith the Bengali digits৯০.style/bn.md§2 asks for Western digits, so the value now reads90. That brings the catalog back to 8 violations, within the limit.The Electron failure came from a merge race. #13883 added a pin that must list every
probeGatewayPortOwnercall and every owner-sentinel literal ingateway-supervisor.js. #13344, merged the day before, had already added two checks the pin does not list:successorOwnsPort), which confirms that the process now holding the port is this app and fails the restart if it is someone else;foreignholder when no remote crew is configured.Both decide who holds the port, not whether it is held. So they are identity judgements, and I added six exact-text entries for them to
IDENTITY_SITES. The occupancy checks in the pin still pass unchanged.Backwards compatibility
Compatible: one Bengali display string and a test list change. No key, schema or runtime behaviour changes.
Tests
website/electron/test/port-owner-self-asserted-identity.test.js: 7 of 7 pass locally, where 2 failed before. Each new entry pins its exact source text and count, so rewording either check fails the pin again.bn.jsongives 8, which is within the limit of 8.Manual verification
N/A. The unit pins are the only consumers of these two changes.
Related Issues
Fixes #14519
Pattern harvest
Rule candidate: review-prompt
Pattern: an exhaustive-list source pin merged after a sibling PR added matching sites; the author should rebase the pin onto current main and re-run it before merge, because CI on a stale base cannot see the sibling's sites.
Checklist
Contribution License Agreement
N/A. The CLA text has not been supplied yet.