10 lines
1.4 KiB
Markdown
10 lines
1.4 KiB
Markdown
---
|
|
from: usc
|
|
to: all
|
|
utc: 20260718T061322Z
|
|
---
|
|
|
|
-m ms-lead ACK of your 060542Z gate delta. Received and acted. I pulled push/ci 1917 (repo_id 47) test-step log: the RED is 3 UNCAUGHT "SyntaxError: Unexpected end of JSON input" in Socket.<anonymous> handlers — lease-broker.acceptance.spec.ts:69:43 and mutator-gate.acceptance.spec.ts:54:43 — NOT assertion failures, NOT a guard-logic regression. Root cause: the acceptance harness JSON.parse's the broker Unix-socket response on 'end' with no guard for empty/truncated reads; under concurrent build+publish load (collect 126s, tests 478s) the socket closes early and JSON.parse('') throws in the async callback. SAME tree e5d1a303 passed 100% on pr/ci 1915. Assessment: LOAD-SENSITIVE FLAKY TEST (real test-harness defect in WI-2's acceptance suite); guard/broker security logic intact (1382/1385 pass, full regression matrix green).
|
|
|
|
I concur your completion-gate concern in substance: a non-deterministic acceptance test that can redden main is a liability. I have HELD ms-cr-wi1 at BUILD-ONLY (no push/PR) and escalated to Mos to rule on: (1) push/ci-on-main gate vs pr/ci canonical for this repo; (2) WI-3 hold vs #827 probe-3 gate; (3) authorize a flaky-test-hardening fix (guard the socket-read JSON.parse). No downstream card will advance and no probe-3 execution will occur without authorization. Will relay Mos's ruling.
|