[steipete]

The runtime reports Node 26.12.1 but did not publish sqlite.db.query. Consumers consequently enabled Node-version-gated diagnostics without receiving the event, including OpenClaw's same-query schema-read reentry test.

Publish through the public diagnostics channel from SQLite's native profile callback. Events are synchronous, including nested queries, with a null-prototype { sql, database, duration } payload: expanded SQL (source fallback), the exact database object and SQLite's nanosecond duration estimate. Subscriber transitions enable/disable tracing; prepare and partial iteration emit nothing, and finalization/GC suppresses events. Subscribers retain responsibility for bounding their own recursive instrumentation, matching Node.

Keep active statements guarded through cleanup resets, reject closing busy statements from authorizer callbacks, and preserve database/publisher lifetime and pending query errors during nested reads. Initialize the lazy next-tick function when tracing activates so stack-overflow errors retain the public channel's asynchronous delivery. Add the current constructor aliases and statement close/disposal support used by Node's unchanged feature tests.

Validation:

• Official Node 26.12.1 source and oracle binaries verified against published SHA-256 checksums. All 12 unchanged Node feature tests pass on Node and the final production build.
• Eight focused regressions pass, including 16-level synchronous nested queries, unguarded RangeError delivery, failed-query cleanup, authorizer safety and subscription transitions.
• Surrounding SQLite and diagnostics-channel suites: 949 passed, with seven inherited platform skips and three inherited TODOs.
• OpenClaw src/infra/sqlite-schema-contract.test.ts at 77d0f72273cb1aaeb21e8e84e6e0742d14de2c03: published S8 24 pass / 1 fail; final patched production build 25 pass / 0 fail / no skips, using isolated HOME/state and OPENCLAW_BUN_PATH.
• Independent Codex review: P2 scoped-clean after resolving cleanup-reset and authorizer-finalization findings; the follow-up type annotation is also scoped-clean.
• Exact-head native CI: Linux x64 and macOS arm64 passed on f5e7fda70b22f27dda67ff59493037d0ef043b91; the CI merge commit has exactly the same source tree. Format, source lint and JavaScript lint/typecheck also passed.

Linux production ThinLTO cost, ten alternating S8/patched pairs pinned to one CPU, 200,000 warmups and 5,000,000 measured prepared SELECT 42 AS answer calls per process:

Measurement	Result
Published S8 median	59.6492 ns/query
Patched, no subscribers	59.3499 ns/query
Median paired ratio	0.99650, gate <=1.02
Ratio of medians	0.99498
Patched, subscribed (ten runs)	262.4756 ns/query
Subscribed / unsubscribed	4.42251x (+203.1256 ns/query)

Every subscribed run verified 5,000,000 events and correct SQL/database identity; all runs verified query results. No timing retries. The small unsubscribed difference is within normal noise, not claimed as a speedup. Baseline is published S8 at 5534724c02; final patched head is f5e7fda70b22f27dda67ff59493037d0ef043b91.

Reference: Node profile callback, subscriber transitions, and unchanged upstream tests. Upstream Bun searches found no matching diagnostics producer to port.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-26T04:53:02.614844Z	f5e7fda	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.