[richiemcilroy]

Why

Every pointer move while dragging the playhead with playback paused rebuilt the whole editor (header, toolbar, sidebar, transport, timeline) twice. seek_to_time notifies the editor once, then the engine echoes the seek back through playhead_changed(frame). seek_to_time had stored the unrounded pointer time while the echo carries frame / 60, so the echo's "already there" check never matched and invalidate_playback_chrome notified the whole editor a second time, a frame later.

What

A paused seek now stores frame / EDITOR_PREVIEW_FPS, the exact value its echo carries, so the echo returns early. Playing seeks are unchanged (the echo there only touches the timeline and transport). playhead_changed and the seek share the same frame_seconds helper so the two values are bit-identical.

The settled playhead is the same as before: it already became frame / 60 once the echo landed. The only difference is that the line is drawn on the frame grid immediately instead of one draw later.

Measured

Scratch A/B on a Blacksmith macOS 26 runner, the GPUI editor driven through the real ruler scrub handlers (timeline_hover then timeline_mouse_move) every 16 ms for 10 s, 3 rounds, 40 s 1080p30 studio fixture.

per scrub step	base	head
full editor rebuilds (header, sidebar, toolbar renders), forward	1.99 / 2.00 / 1.99	1.00 / 1.00 / 1.00
full editor rebuilds, backward	1.99 / 1.99 / 1.99	1.00 / 1.00 / 1.00
main thread CPU, forward scrub	35.5% / 40.7% / 41.1%	24.8% / 24.0% / 23.8%
main thread CPU, backward scrub	40.3% / 42.7% / 34.1%	13.3% / 24.4% / 19.8%
seek to frame on screen p95, forward	7.9 / 8.5 / 9.5 ms	4.8 / 4.8 / 5.3 ms

Frames presented per seek are unchanged (1.00 forward on both). Playback and hover counters are unchanged: no section other than the timeline and transport re-renders during playback on either side.

Tests

paused_seek_lands_where_its_echo_does checks, for every frame of an hour at 60 fps, that the paused seek stores exactly the time playhead_changed computes for the echo (so its early return fires), and that a playing seek keeps the pointer time.

Risk

Low. Every section still rebuilds once per paused seek, so nothing that reads the playhead can go stale.

Confidence Score: 5/5

The change appears safe to merge.

Summary

Paused seeks now store the requested frame’s time immediately, so the matching engine echo does not trigger another editor rebuild.

• Paused playhead seeks now avoid a second full-editor rebuild.

No actionable issues were found. Tests were inspected, not run.

_{Reviews (1) · Last reviewed commit: "perf(desktop-gpui): store the frame time..." · Reviewed by Greptile}

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
cap-status	Ready	Preview	Oct 22, 2026 12:29am UTC

[blacksmith-sh]

Found 1 test failure on Blacksmith runners:

Failure

Test	View Logs
cap-verify::crash/crash_safety	View Logs

^{Need help on this PR? Tag @codesmith-bot with what you need.}

[richiemcilroy]

@greptileai review

[capy-ai]

Capy found no issues.

View 1 other finding in Capy.