Repository navigation
Description
Problem and impact
The existing long-history regression cannot page to the start and verify every message part. This blocks V2 browser verification and prevents confidently merging the memory-retention work. Investigate whether the cause is timeline behavior or a stale fixture assumption; the current evidence establishes a reproducible regression-test failure, not message loss in an installed app.
Priority: medium correctness investigation and a CI integration blocker.
Reproduction and evidence
Reviewed host base dafdcd4171466050c984d524ca0d1a09d31f100e and CI repair head 65b4da5b81bad5599741416e86d334289f6fc2ac. Neither CI repair commit changes timeline implementation or this test.
From packages/app, with dependencies and Playwright Chromium installed:
PLAYWRIGHT_BUILD=1 PLAYWRIGHT_PORT=4391 bunx playwright test e2e/regression/timeline-history.spec.ts -g "mounts every part once and in order" --workers=1 --retries=0Expected: the production timeline moves through older pages and the fixture observes every expected message and part in order, without errors or gaps.
Actual: the poll at timeline-history.spec.ts:127 times out after 10 seconds waiting for a pending older-page request or a changed, covered visible signature. The test does not reach the final complete-history assertions.
- Hosted Linux job: the full browser step completed with 444 passes and four failures, including this test. Its later component-test phase was cancelled after a newer CI head was published.
- Independent local Linux reproduction against a production build, one worker and no retries, also fails at the same predicate. The focused five-test run finishes four pass / one fail. SSH reconnect, held side-panel transitions, and video fullscreen passed in that focused run, so those separate CI failures are not attributed to this defect.
- The local worktree additionally corrects an unrelated stale effort-picker expectation in composer.spec.ts; timeline source and this test remain unchanged.
Acceptance criteria
- Identify and document whether the failure belongs to production paging/virtualization or fixture synchronization.
- The fixture reaches the start and observes every expected message and part exactly once and in order.
- Preserve visible-row anchoring when older pages prepend and preserve existing error checks.
- Fix the cause without skipping the scenario, reducing its fixture coverage, or merely increasing its timeout.
Verification
Retain the existing regression. Reproduce before the fix, run it after the fix, and run the neighboring history/retention cases against a production build. Verify the corrected head in hosted Linux and Windows CI. The actual installed CM was not restarted or modified by this investigation.
Related: #77, #87. This is separate from backend transcript eviction in #77.
Activity
fix(cm): integrate reviewed V2 reliability fixes [skip ci]
trmosala commentedon Oct 21, 2026
Merged into
CM-Split-v2as290d7d6353c0c7a4a356af780b5687f2dcfe5cfcthrough PR #96. Its tree exactly matches reviewed source68ab9205709350385471eb2d383b3cff3c18e678.Implemented in PR #96, reviewed source
68ab9205709350385471eb2d383b3cff3c18e678.The production timeline now reads live element offsets while a scroll is active. The retained history regression reaches the beginning and verifies every expected message/part once in order, while retaining anchor and error checks. Paging coverage was not skipped or reduced.
Final verification is deferred at the user's explicit request to finish and close the implementation issues before one final test pass. This closure records implementation completion, not a claim that the final combined/platform checks have passed. Any failure in that pass will be recorded against the relevant issue.