Repository navigation
fix: timeout synthetic probe invoke - #62
Conversation
|
🦞👀 Pull request received. I will update this pull request when review starts. ClawSweeper review completeClawSweeper finished reviewing this revision. The review result is being finalized. |
|
Codex review: blocked before merge. Reviewed September 4, 2026, 12:39 PM ET / 16:39 UTC. ClawSweeper reviewWhat this changesThis PR adds configurable deadlines to synthetic hook and registration invocations so never-settling handlers are reported as failed probes instead of blocking the runner. Regression provenancePossible regression — probable (reviewed change; failure trace). No predecessor PR is attributed. Merge readiness⛔ Blocked before merge - 6 items remain Keep open: the timeout addresses a source-proven hang and has direct after-fix terminal evidence, but it introduces a new 30-second failure default and public timeout controls without an approved compatibility contract or the required Crabpot consumer smoke. Priority: P2 Review scores
Verification
How this fits togetherPlugin Inspector captures plugin hooks and registrations, then invokes synthetic inputs to produce an inspection report. This change sits in the in-process probe runner between captured handlers and the resulting pass/fail records. flowchart LR
A[Plugin entrypoint] --> B[Captured hooks and registrations]
B --> C[Synthetic probe runner]
C --> D[Invocation deadline]
D --> E[Pass or fail result]
E --> F[Inspection report]
Decision needed
Why: This changes established execution semantics for existing users and exposes new package configuration; source evidence cannot establish the intended compatibility policy. Before merge
Findings
Agent review detailsSecurityNone. Review metrics
Merge-risk optionsMaintainer options:
Technical reviewBest possible solution: Settle whether bounded synthetic execution and its override controls are a stable supported contract, then document and test slow-success/override behavior and validate the exact head through Crabpot. Do we have a high-confidence way to reproduce the issue? Yes: current main directly awaits probe invocation with no deadline, and the supplied post-fix terminal trace exercises the public captured-hook path with a hanging Promise. Is this the best way to solve the issue? No: racing the invocation is a targeted mechanism, but the forced 30-second default and newly public controls need an explicit compatibility decision plus slow-success and override validation. Full review comments:
Overall correctness: patch is incorrect AGENTS.md: found and applied where relevant. Codex review notes: model internal, reasoning high; reviewed against 92db8c57e1d5. LabelsLabel changes:
Label justifications:
EvidenceWhat I checked:
Likely related people:
Rank-up movesOptional improvements that raise the rating; they are not merge blockers.
Rating scale
Overall follows the weaker of proof and patch quality. Workflow
HistoryReview history (11 earlier review cycles; latest 8 shown)
|
e3be165 to
92db8c5
Compare
Signed-off-by: Sebastien Tardif <SebTardif@ncf.ca>
Integrate main while preserving the contributor repair and retained callback identity. Add malformed-report regressions for the approved protocol-validation follow-up; these cases intentionally establish the red CI baseline.
Validate report shape, row identities and statuses, and exact safe-integer counters before artifact publication. Preserve empty and completed failed or blocked reports. This is protocol validation, not same-process authentication.
What Problem This Solves
Fixes an issue where users running synthetic probes could hang on callbacks, continue dependent work after a timeout, or leave plugin code running after the helper CLI should have finished.
Why This Change Was Made
Keep programmatic probes in-process with finite callback deadlines, cooperative cancellation, and retained callback identity. Run CLI capture and callbacks together in the existing supervised capture runner, preserving its mock-SDK loader lifetime, bounded plugin output, and complete report delivery.
The CLI reads a fresh report separately from plugin stdout. The existing regular-file, nonblocking, byte-bounded reader is shared with import-loop profiling, so a FIFO or oversized artifact cannot bypass the child boundary. No timeout constants or resolver helpers are added to public exports.
User Impact
PLUGIN_INSPECTOR_PROBE_TIMEOUT_MS, then 30 seconds. Zero, infinite, malformed, and overflowing values do not disable limits.Evidence
bf4238e8e2e1a53e832559d744a358990a069cc3and integrates main340f18e26558c5f6530d322ac399b76b25e6b267, retaining the landed child supervisor, real capture policy, serial lifecycle order, and CJS loader changes.c59d84a1bf3f08104a259d7d2c7c4026b763f7b3. Protocol red head295746581c7381567e58aa9e6f07157cf62b7321kept that production unchanged: secretless Node 22 Check 34376448341 produced exactly 52 expected malformed-report rejection failures and 389 passes, including the valid-report controls.6aa4aff1336005a75014b449b97e0aef6d129f60; tree:145fd0a66a21491b44db772fed2d4f85889d6053. Exact-head secretless Node 22 Check 34376704253 passed all 441 tests with zero failures, cancellations, or skips, plus package-content validation (job102551213467).The maintainer approved the finite 30-second default and override contract. The original test-only timeout exports were removed. No new public timeout resolver or authentication mechanism is introduced. Exact-head CI is green; final maintainer review and squash-merge clearance were granted for
6aa4aff1336005a75014b449b97e0aef6d129f60.Linux owned-child proof only; Windows keeps the existing direct-child cleanup contract. Gateway response validation is deferred to its separate unit. No dependency, package version, or release change. Crabpot followthrough remains assigned to release integration.
Thanks @SebTardif for the original timeout repair.