[Hackathon] BE-closure fan-out in "rerun --wait" aborts the batch and detaches sibling polls

[Andy00L]

Discord Username / User ID

interferon0

What does this improvement do?

On single-test testsprite test rerun --wait for a backend test whose dependency closure has 2+ members, if any member's poll throws a non-timeout error mid-flight (a 200 with a non-JSON body, or a transient AUTH_INVALID/NOT_FOUND/RATE_LIMITED), .catch(reject) rejects the whole fan-out: every already-collected result (including the named test's passed verdict) is discarded and the sibling polls are left detached with open fetches and timers. The two --all pools were fixed for exactly this failure; the closure fan-out was not.

Details / implementation notes

src/commands/test.ts:6143-6148 (pollMember rethrows any non-TimeoutError) and :6195 (.catch(reject)), with :6201-6228 only special-casing RequestTimeoutError. Mirror the fixed sibling pools (pollFreshAccepted ~5696, pollAccepted ~6897): make pollMember total (map ApiError/timeout/raw error to a captured result instead of throw) and replace .then(...).catch(reject) with the same recordSettled helper, plus a shared closure deadline so each member gets the time REMAINING rather than a fresh full --timeout. Note this is adjacent to the filed rerun exit-code work (#78/#33) but is a distinct code path and defect (fan-out totality + the named verdict being lost). No new dependency. Tests: a member's transient error is captured while siblings finish and the named passed verdict survives. PR to follow once assigned.

Confirmations

• I have searched existing issues and this is not a duplicate.
• I have provided my Discord identity above for reward coordination.

[zeshi-du]

Triage: accepted (this one was missed in the 07-03 batch — your three sibling reports all got verdicts and this didn't; apologies). Re-verified against current main: pollMember still rethrows non-timeout errors and the fan-out still rejects wholesale via .then().catch(reject), exactly as described — the two --all pools got the equivalent fix (#154/#130) but this closure fan-out path didn't. Now also tracked internally. A PR following the #154 pattern (classify per-member, collect partial results, surface failures in closureFailures[]) is welcome; comment /assign if you want it.

[testsprite-hob]

Assigned to @zeshi-du. Thanks for taking this on. ⚠️ You're now over the soft 3-issue limit (4/3). Consider finishing one before taking more.

Your open assigned issues (4/3):

• MSW contract suite validates an obsolete LatestResult schema #255 — MSW contract suite validates an obsolete LatestResult schema
• Runtime Node guard accepts releases excluded by package engines #254 — Runtime Node guard accepts releases excluded by package engines
• [Hackathon] Width-aware progress ticker #72 — [Hackathon] Width-aware progress ticker

[zeshi-du]

Fixed on our side: rerun --wait's closure fan-out now classifies per-member poll errors instead of aborting the batch, so one member failing no longer detaches its siblings' polls. It's in the current release.

The fix landed internally rather than through a public PR, which is why this issue stayed open with nothing visibly happening — that's a gap in how we close the loop, and it's the second issue in this sweep that sat open purely for that reason. Closing.

Thanks for the precise write-up of the failure mode; it made this a small fix.