perf(recording): take the cursor position from the recorder's mouse sample

[celeroncoder]

spawn_cursor_recorder keeps one DeviceState and reads get_mouse() from it every tick (about 60/s) for the buttons. It then called RawCursorPosition::get() for the position, which is DeviceState::new().get_mouse().coords. So every tick built a second DeviceState and queried the pointer twice.

What that extra DeviceState::new() costs:

• Linux (X11): XOpenDisplay + XCloseDisplay, so one X connection set up and torn down per tick.
• macOS: device_query assert!s application_is_trusted_with_prompt(), so the Accessibility trust check ran every tick.
• Windows: a second GetCursorPos plus the button polls.

Change

• Adds RawCursorPosition::from_coords (crates/cursor-capture/src/position.rs). get() calls it and keeps working for the cap-cursor-capture binary.
• spawn_cursor_recorder builds the position from the MouseState it already has, in all three places: the initial sample, the waiting-for-start-gate branch and the per-tick sample.

 let mouse_state = device_state.get_mouse();
-let position = cap_cursor_capture::RawCursorPosition::get();
+let position = cap_cursor_capture::RawCursorPosition::from_coords(mouse_state.coords);

Positions come from the same source as before. On Linux both handles run XQueryPointer on the default root of XOpenDisplay(NULL). On macOS and Windows DeviceState is a unit struct and the position is readmouse::Mouse::location() / GetCursorPos. The only difference is that position and buttons now come from one sample instead of two taken microseconds apart.

Results (Linux X11)

	main @ 83237e9	This PR @ b4f88a6
X11 connect() per tick, moving pointer	2.00 (926 over 463 ticks)	1.00 (504 over 504 ticks)
Recorder CPU per 10 s, idle pointer, 24 px ARGB cursor	259 ms [240-270]	166 ms [160-185]
Recorder CPU per tick, moving pointer, 24 px ARGB cursor	414 µs [407-470]	266 µs [265-269]
Xvfb CPU per 10 s, idle pointer, 24 px ARGB cursor	240 ms [220-260]	140 ms [140-150]

The real spawn_cursor_recorder ran for 10 s from release builds of both commits against Xvfb 1920x1080 on a shared 2 vCPU aarch64 VM. CPU is the recorder's user+sys from getrusage, median of 5 interleaved runs with the range in brackets. Connection counts come from one strace run, with the startup connections subtracted.

• Most of the saving is sys time. At 24 px idle, recorder sys went from 168 to 91 ms per 10 s and user from 91 to 75 ms.
• The idle runs shared the VM with a background build (load1 1.13-3.74). The moving runs were at load1 0.20-0.96.
• Xvfb is a local server with no compositor, so a real X.Org or XWayland session will show different absolute numbers.
• With the pointer stepped through 8 fixed positions, both builds recorded the same 8 pixel positions and the same cursors.

macOS and Windows weren't built or measured. The code there is the same platform-neutral change, and on macOS it removes the per-tick Accessibility check.

Overlap

• feat(linux): enable Wayland window/display recording with PipeWire cursor metadata and evdev clicks #2420 (Wayland) replaces last_mouse_state with last_buttons. If both land, the two last_position lines become RawCursorPosition::from_coords(device_state.get_mouse().coords).
• perf(cursor): stop hashing an unchanged cursor image every tick #2466 doesn't touch these lines.
• perf(windows): fence-waited readbacks, overlapped MF export decode, cheaper screen capture #2541, refactor(rendering): drive the Media Foundation decoder from one request state machine #2545 and perf(video-decode): pipeline the Media Foundation GPU readback #2583 add a Windows cursor-image cache in the same file. Their cursor.rs hunks apply cleanly on top of this.

How this was made

This PR is agent-generated.

• Gemini 4 Argon (Antigravity) and space-bunny-free (OpenCode) found this during an audit of the repo and wrote the first patch.
• Claude Opus 5.5 reviewed that patch, rewrote it on current main and took the measurements above with a probe that drives the real recorder.
• Gemini 4 Argon reviewed the final diff and numbers over two rounds and approved it.

🤖 Generated with Claude Code

Confidence Score: 5/5

The PR appears safe to merge; no actionable issues were found.

Summary

The recorder now takes cursor coordinates from its existing mouse sample instead of creating another DeviceState.

• The cursor recorder gets positions from its existing mouse sample.

_{Reviews (1) · Last reviewed commit: "perf(recording): take the cursor positio..." · Reviewed by Greptile}

[vercel]

@celeroncoder is attempting to deploy a commit to the Cap Software Inc Team on Vercel.

A member of the Team first needs to authorize it.