desktop: timeline rows lack containment, so scrolling a long session burns main-thread time in layout/paint #54291

Activity

1. github-actions

   assigned

   Hona

   on Oct 10, 2026
2. opencode-agent commented on Oct 10, 2026

   

   opencode-agentBot

   on Oct 10, 2026 – with OpenCode Agent

   Contributor

   More actions

   Thanks for the detailed measurements. I tried to look at this here, but my sandbox is a headless, single-core Linux machine with no GPU compositor and no desktop app. Frame-rate and layout/paint timings from it wouldn't be comparable to yours, and with a metric this noisy they also wouldn't be stable enough to compare paired runs. A maintainer will need to run it on real desktop hardware.

   What I could confirm from the code: the row style in packages/app/src/session/timeline/virtualizer.tsx sets only overflow: clip / overflow-clip-margin, and packages/app/src uses no contain or content-visibility, as you describe.

   One thing that would help: I couldn't find session-timeline-scroll-benchmark.spec.ts in the repository (the closest is packages/app/e2e/performance/timeline/session-timeline-benchmark.spec.ts). If you wrote the scroll benchmark yourself, could you share it, or the exact command you ran, so maintainers can repeat your paired runs?

   Related, but not the same problem:

   • [Desktop] Session panel width animation costs ~0.24s of main-thread work per toggle #48701: another desktop main-thread cost, from the session panel width animation
   • refactor(app): avoid deep equality in timeline row reconciliation #48435: an open PR that reduces timeline row reconciliation work (JS side, not layout/paint)
3. Te-River commented on Oct 10, 2026

   

   Te-River

   on Oct 10, 2026

   Author

   More actions

   The benchmark is in #54293 — packages/app/e2e/performance/timeline/session-timeline-scroll-benchmark.spec.ts plus its probe. Run it from packages/app with PLAYWRIGHT_BUILD=1:

   node_modules/.bin/playwright test --config e2e/performance/playwright.config.ts timeline/session-timeline-scroll-benchmark.spec.ts --repeat-each=20 --workers=1 --retries=0
   

   Two notes for whoever runs it on desktop hardware:

   • The frame-gap metric is bimodal (50ms / 66.6ms) and the mode tracks machine load, so compare paired/interleaved runs inside the same window rather than absolute numbers across windows. That is what the numbers in the PR come from.
   • The two scenarios bracket the cold-tail settle: one scrolls as soon as the timeline mounts, the other waits for the settle to finish.

   Thanks for confirming the row style from the code.