Verify Windows backend readiness failures and repair standalone memory smoke #1374

Activity

1. currentbits

   added

   plan:todoPlanboard statusPlanboard status

   plan:doingPlanboard statusPlanboard status

   and removed

   plan:todoPlanboard statusPlanboard status

   on Sep 28, 2026
2. currentbits commented on Sep 28, 2026

   

   currentbits

   on Sep 28, 2026

   OwnerAuthor

   More actions

   Progress:

   • Fresh run 36327348080 at 6f9a305: all six jobs passed; neither failure recurred there.
   • Codex readiness already fixed by 1a19ebb in merged Fix MCP discovery, approvals, and provider skill selection #1370, using actual injected handle.canSteer() rather than fake-server request recording. All six checks passed in run 36387517093. No duplicate Codex or production steering changes.
   • On separate branch fix/windows-backend-readiness (base fc354cc), forced a real occupied port in the standalone memory smoke. Original helper reproduced the 8000ms timeout (8054ms total). Following the persisted fallback port and verifying nonce proof passed the same case in 127ms.
   • Child close is registered immediately; test cleanup awaits pipe closure and uses existing asynchronous rmTree. Added early-child-exit diagnostic/close regression. No timeout increase or production changes.

   Historical Windows root cause remains unprovable from that log because child output was not collected. The port-selection mismatch is independently reproduced, not inferred solely from the timeout. Final verification and review annotation in progress.
3. currentbits commented on Sep 28, 2026

   

   currentbits

   on Sep 28, 2026

   OwnerAuthor

   More actions

   Local verification complete: Node v22.23.2, all inherited CODER_* variables removed, npm test --prefix memory-server: 234 passed, zero failed/skipped. git diff --check clean. Review itinerary written locally at .solenta/review-itinerary/0ec40e21-bfa4-466c-93e2-6683f30dec41.json (repository-ignored).

   Committed test-only repair 9b15dad on separate branch fix/windows-backend-readiness; pushed for native CI: https://github.com/currentbits/solenta/actions/runs/36406318495 . Waiting for hosted results before final completion. PR #1367 and production code are untouched.
4. currentbits

   added

   plan:donePlanboard statusPlanboard status

   and removed

   plan:doingPlanboard statusPlanboard status

   on Sep 28, 2026
5. currentbits commented on Sep 28, 2026

   

   currentbits

   on Sep 28, 2026

   OwnerAuthor

   More actions

   Completed the scoped investigation and separate repair in draft PR #1375, branch fix/windows-backend-readiness, commit 9b15dad. All plan steps finished; no merge performed.

   Fresh run 36327348080 at6f9a305a was fully green. Codex steering readiness already repaired in merged #1370 by waiting for actual canSteer; no duplicate change here. Memory forced-collision reproduction failed with the original 8000ms wait, then passed after following persisted port and checking health proof. Child output is drained/captured, early failures are diagnosed, and cleanup waits for pipe close then existing async rmTree. Production code and existing Windows assertions unchanged.

   Native verification: https://github.com/currentbits/solenta/actions/runs/36406318495 — Windows and Linux full tests plus all three smoke jobs passed. Windows steering passed (779ms), forced memory collision passed (1999ms), early-exit/pipe-close test passed (55ms). Local Node22.23.2:234/234 memory tests passed. macOS also passed234/234 memory tests, but overall macOS job failed at unchanged electron/test/eject-command.test.js:161 (copied.includes('-m') true). That out-of-scope failure was suggested as separate work; PR remains draft and overall CI is not green.

   Historical Windows memory trigger cannot be proven from its missing child output; the readiness defect is deterministically reproduced. #1367 untouched; package validation stays in #1366. Review itinerary updated with final CI limitation.
6. currentbits

   closed this as completedon Sep 28, 2026