Canvas: zoom regression with 21 windows (`678b4156` vs `d827b3e`) #124

New issue

Closed

opened 2026-10-09 06:31:00 +00:00 by kayg · 3 comments

kayg commented

2026-10-09 06:31:00 +00:00

Owner

See main 0bdfaae9 (DECISIONS/TASTE) and the owner's report of 2026-10-09.

See main 0bdfaae9 (DECISIONS/TASTE) and the owner's report of 2026-10-09.

kayg commented

2026-10-09 08:17:52 +00:00

Author

Owner

Finished (head c4acc5f7). Bisect of d827b3ed, 98f6f94a, a9ae1435, 8413a44c, 678b4156 (3 s ctrl-wheel zoom, scene 7, 21 windows, software raster, 4 alternating runs): p50 16.7 ms on every commit, renderer+GPU CPU 5.3-5.7 s on every commit; none of the commits touches the camera or LOD. The owner's slow sessions ran 32 px card blur; Frosted windows double the zoom CPU (9.96 s vs 5.18 s). Fix 932b814c: backdrop blurs off while the camera moves (data-camera-moving).

Finished (head c4acc5f7). Bisect of d827b3ed, 98f6f94a, a9ae1435, 8413a44c, 678b4156 (3 s ctrl-wheel zoom, scene 7, 21 windows, software raster, 4 alternating runs): p50 16.7 ms on every commit, renderer+GPU CPU 5.3-5.7 s on every commit; none of the commits touches the camera or LOD. The owner's slow sessions ran 32 px card blur; Frosted windows double the zoom CPU (9.96 s vs 5.18 s). Fix 932b814c: backdrop blurs off while the camera moves (data-camera-moving).

kayg commented

2026-10-09 08:26:58 +00:00

Author

Owner

Real-GPU A/B on the macOS VM (Chromium, DPR 2, 1600×1000, scene 7 = 21 windows, theme dracula + Mesh background, Frosted and Solid, 4 alternating rounds, 3 s ctrl-wheel zoom): d827b3ed p95 19.7–23.4 ms, max 23–149 ms; 678b4156 p95 18.8–21.6 ms, max 21–110 ms. No regression between the builds; the VM display is 60 Hz so it cannot show 120 Hz dips. Zoom cost at 21 windows is the real problem → tracked under #121 with an 8 ms work-per-frame budget measured as work time, not frame intervals. The mitigation 932b814c (blur off during camera moves) is being reverted (owner rule: never cheapen; glass never shows legible text).

Real-GPU A/B on the macOS VM (Chromium, DPR 2, 1600×1000, scene 7 = 21 windows, theme dracula + Mesh background, Frosted and Solid, 4 alternating rounds, 3 s ctrl-wheel zoom): d827b3ed p95 19.7–23.4 ms, max 23–149 ms; 678b4156 p95 18.8–21.6 ms, max 21–110 ms. No regression between the builds; the VM display is 60 Hz so it cannot show 120 Hz dips. Zoom cost at 21 windows is the real problem → tracked under #121 with an 8 ms work-per-frame budget measured as work time, not frame intervals. The mitigation 932b814c (blur off during camera moves) is being reverted (owner rule: never cheapen; glass never shows legible text).

kayg commented

2026-10-28 17:52:43 +00:00

Author

Owner

Closed by the orchestrator during the 2026-10-28 handover (#254), on the owner's request to close finished issues. Evidence: Bisect found no regression across the commits; real zoom cost is tracked under #121 and the blur mitigation was reverted. Reopen if anything is still missing.

kayg closed this issue

2026-10-28 17:52:43 +00:00

~~kayg referenced this issue 2026-10-28 17:53:54 +00:00~~

Orchestrator handover 2026-10-28: moving to a new machine #254

kayg referenced this issue

2026-10-28 18:00:45 +00:00

Orchestrator handover 2026-10-28: moving to a new machine #254