Measure the ring with live ZBrush: hotkey to pixels, pick to brush changed, picks mid-stroke #213

The ring performance audit of November 2026 measured the renderer and the daemon off screen (#210 came out of it). Nothing was measured with live ZBrush, the real keyboard hook or a pen. These are the measurements still to take, each with how.

Hotkey to pixels, end to end

• In a scratch branch, stamp perf_counter at the hook callback (compare KBDLLHOOKSTRUCT.time with GetTickCount), at the WM_APP_OPEN dispatch, after the first UpdateLayeredWindow and after ShowWindow. A 240 fps phone video of key press to ring visible is the outside check.
• For the first open after start and for later opens, before and after The ring opens faster: pictures kept between opens and drawn ahead of the first #210.

ZBrush's frame time while the ring is up

• PresentMon on ZBrush with the ring idle, hovering and in its open animation, at 100% and at 150% scaling. The ring is a large topmost window with per-pixel alpha.

Release to brush changed

• radiald.log has millisecond stamps ("closed (released)", "pick #N -> listener"). The bridge's log has whole seconds, so log milliseconds there for the run, or time apply_request in ZBrush's Python console.
• Separately: the existence check, the press, the extra read of Transform:Edit, and a brush loaded from a file, which goes through the 40 ms select retry (up to 1 s).

A pick during a stroke

• The listener polls for the button on a 40 ms timer, which may come out near 47 ms at the default timer resolution. Measure button-up to brush changed; a 16 ms timer may be worth it.

ZBrush 2022 to 2025

• The bridge-key route: key sent, ZScript reads the request file. Also how often the 1 s and 6 s fallbacks fire in ordinary use.

The app while idle

• CPU sampling of the app over five minutes of sculpting with ZBrush in front. See the issue on process snapshots for what to expect.
• The count of "hook re-installed" lines in radiald.log over a long session.
• Memory over a day of use. The audit saw a drift of about 1 MB per 100 opens, before The ring opens faster: pictures kept between opens and drawn ahead of the first #210.

The browser window, on the pen's path

• While the browser window is visible it registers raw mouse input with INPUTSINK (zorbit/popout.py), so every pen or mouse report becomes a WM_INPUT handled in Python. Compare its CPU while sculpting with the browser shown and hidden.

After #210

• The ring warm-up line in radiald.log gives the time and the longest step on a real setup. Check it after start and after a save from Studio, on a monitor at the scaling actually in use.