Details
- Reviewers
npodgurski cmurray jlewis - Group Reviewers
ai-security-reviewers - Bugzilla Bug ID
- 2064487
- Merge Conflict Status
- Landing merges cleanly into the target branchMerges cleanly when merged against target branch tip 6210f73af7ef, starting from 751f5671de2d.Last checked Fri, Oct 16, 9:29 PMDiff 1410099, based on 751f5671de2d
Diff Detail
- Repository
- rFIREFOXAUTOLAND firefox-autoland
- Branch
- HEAD
Event Timeline
(Reviewing Diff 1409468)
The schema change, migration and set-union upsert look correct: NULL values are coalesced, and the new column is guarded by both the version check and a column-existence check. The scheduler change follows the same metadata shape as the midChat path. The main weak spot is cleanup timing in the new browser test.
| browser/components/aiwindow/models/tests/browser/browser_llmaj_telemetry.js | ||
|---|---|---|
| 91 | markStub.called turns true while #onInterval is still running. After markLLMTelemetryProcessed the loop waits 3–4 s on setTimeout, and only then does its finally run and set LAST_RUN_PREF to Date.now(). Calling destroy() doesn't cancel that pending run. So the run keeps going after the sandbox is restored, and it overwrites the pref this block restores once the task has finished. That leaks state into later tests. Fix by waiting for the run to finish before cleanup. One way: wait for LAST_RUN_PREF to become non-zero. Another: stub the delay so it resolves right away. | |