Perf harness needs a launch-time selection to measure scroll-spam #34

New issue

Open

opened 2026-09-16 01:21:17 +00:00 by kayg · 0 comments

kayg commented

2026-09-16 01:21:17 +00:00

Owner

Overnight (Main): Scripts/perf/perf-app-trace.sh / perf-scroll-signposts.sh cannot produce accepted evidence on the VM fixture Space: the app launches with "Nothing selected", so ViewerHostingView.draw never marks viewer content and startInputDriverIfRequested never runs (report: "missing real viewer content draw", "Scroll was not measured") — with or without -synthetic. The repeat harness needs a launch-time selection (restore last selection per Space, or a CALTERNAL_PERF_SELECT_FIRST_DAY bootstrap).

What was measured for the morning build (Release, VM VirtualMac2,1, macOS 26.6.2, fixture "Calternal Overnight Perf QA"): process-start → first timeline content draw, 3 alternating runs each — baseline Calternal-dev-release (10768c67): 647 / 649 / 641 ms; merged desktop-test (f775ed08): 608 / 610 / 586 ms. No launch regression (VM is above the 500 ms gate for both; gate is judged on real hardware). Key-repeat: a 15 s sample during a Down-key walk on the Release merged build shows the main thread 88 % idle in mach_msg, no thumbnail/TextKit/geometry frames — but cua can only deliver ~2.6 presses/s, so this is not key-repeat-rate proof.

Overnight (Main): `Scripts/perf/perf-app-trace.sh` / `perf-scroll-signposts.sh` cannot produce accepted evidence on the VM fixture Space: the app launches with "Nothing selected", so `ViewerHostingView.draw` never marks viewer content and `startInputDriverIfRequested` never runs (report: "missing real viewer content draw", "Scroll was not measured") — with or without `-synthetic`. The repeat harness needs a launch-time selection (restore last selection per Space, or a `CALTERNAL_PERF_SELECT_FIRST_DAY` bootstrap). What was measured for the morning build (Release, VM VirtualMac2,1, macOS 26.6.2, fixture "Calternal Overnight Perf QA"): process-start → first timeline content draw, 3 alternating runs each — baseline `Calternal-dev-release` (10768c67): 647 / 649 / 641 ms; merged desktop-test (f775ed08): 608 / 610 / 586 ms. No launch regression (VM is above the 500 ms gate for both; gate is judged on real hardware). Key-repeat: a 15 s `sample` during a Down-key walk on the Release merged build shows the main thread 88 % idle in `mach_msg`, no thumbnail/TextKit/geometry frames — but cua can only deliver ~2.6 presses/s, so this is not key-repeat-rate proof.

No labels

No milestone

No project

No assignees

1 participant

Notifications

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

kayg/calternal-macos#34

No description provided.

Rows
Columns