[OptimistOtaku]

Playback and export call cursor_visibility_at for every frame. When a project contains hidden-cursor clips, it constructs and drops a vector of every hidden run just to select the run containing that frame.

Merge runs while retaining the first matching window and the total clip duration, using fixed stack storage. The complete traversal remains necessary for fades at the timeline's end. Fullscreen-hold mapping, transition arithmetic, adjacent-run merging, exclusive end boundaries, and fade calculations retain their previous behavior. The small no-hidden-clips entry path is kept separate from the heavier implementation.

Measurements

Windows x86_64, Ryzen 5 5600H, Rust 1.88 GNU, repository release profile (opt-level=s, thin LTO). Nine alternating paired samples after warmups, each with 10,000 varying frame times. The ignored benchmark compares the exact previous functions with production code and checks complete float-bit equivalence before timing.

Clips	Hidden clips	Transitions	Holds	Before (ns)	After (ns)	Speedup
2	None	No	No	3.33	3.07	1.08×
2	All	No	No	58.01	17.94	3.23×
2	Mixed	No	No	69.65	20.46	3.40×
2	Mixed	Yes	No	75.80	25.36	2.99×
2	Mixed	Yes	Yes	141.38	94.25	1.50×
20	None	No	No	9.82	15.37	0.64×
20	All	No	No	128.95	83.50	1.54×
20	Mixed	No	No	196.30	88.34	2.22×
20	Mixed	Yes	No	305.81	203.31	1.50×
20	Mixed	Yes	Yes	530.48	434.08	1.22×
200	None	No	No	90.84	108.05	0.84×
200	All	No	No	892.96	790.40	1.13×
200	Mixed	No	No	1,176.43	848.08	1.39×
200	Mixed	Yes	No	3,116.00	2,863.57	1.09×
200	Mixed	Yes	Yes	5,931.31	5,729.84	1.04×
1,000	None	No	No	568.59	616.30	0.92×
1,000	All	No	No	5,378.99	4,777.59	1.13×
1,000	Mixed	No	No	5,660.08	4,880.92	1.16×
1,000	Mixed	Yes	No	21,384.78	20,603.92	1.04×
1,000	Mixed	Yes	Yes	40,407.80	42,132.87	0.96×

The benefit is concentrated in hidden-clip projects without large transition/hold workloads. No-hidden controls cost approximately 6–48 ns more in the larger fixtures; the 1,000-clip transition/hold control was about 4% slower. A second complete run retained the hidden-clip gains and that largest control's regression. Hold-window allocation and full clip traversal remain. These measure cursor visibility CPU work, excluding GPU rendering and overall FPS.

cargo test --locked -p cap-project --release --lib benchmark_cursor_visibility -- --ignored --nocapture

Validation

• cargo fmt --all, cargo check --locked -p cap-project
• cargo test --locked -p cap-project --lib --tests: 286 library tests and four integration tests passed after rebasing; seven benchmarks ignored by default. The cursor benchmark measurements above are from the original PR run.
• cargo clippy --locked -p cap-project --all-targets -- -D warnings
• git diff --check

Differential tests cover every hidden/visible pattern through six clips, fractional speeds, transitions, overlapping fullscreen holds, exact fade/window boundaries and neighboring representable timestamps, empty timelines, degenerate clips, and non-finite inputs. No dependencies, platform-specific code, persistent caches, or project-format changes. The native GPUI app was not launched locally. This branch is independent of #3436. Rebased onto main 6fbbaa577; the conflict resolution retains main's marker exports alongside the cursor test module. An existing shape-test field assignment was moved into its initializer to pass all-target Clippy without changing test behavior.

Confidence Score: 5/5

The rebased PR appears safe to merge, with no actionable issues found.

Summary

Removes the hidden-run vector from cursor_visibility_at while keeping the previous window selection and fade calculations.

• Cursor visibility finds the matching hidden run without building a list.
• The shape test sets its infinite width when it builds the shape.

No actionable issues found. Build and test results were inspected as reported, not rerun.

_{Reviews (2) · Last reviewed commit: "test(project): initialize non-finite str..." · Reviewed by Greptile}

[vercel]

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

A member of the Team first needs to authorize it.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has insufficient remaining quota.

An organization owner can choose the repository organization as the billing source for reviews requested by members with a paid Copilot license. Learn about Copilot code review billing.

[richiemcilroy]

Thanks for this. The win is tens of nanoseconds per frame against a multi-millisecond frame budget, and several rows in your own table get slower (no-hidden 20/200/1000 clips, 1000 clips with holds), so it isn't worth the extra code in cursor_visibility_at.

[OptimistOtaku]

Thanks Richie, that's fair. I put too much weight on the relative microbenchmark gains here. Removing the allocation still leaves the full clip traversal and hold mapping, and the regressions in the no-hidden and large hold cases weaken the case further.

I agree the absolute savings don't justify the added complexity in this per-frame path. I'll take that into the next contribution: profile a meaningful bottleneck first, then check the absolute savings and common-case controls before proposing the change.

[richiemcilroy]

Thanks! Can you rebase on main to clear the conflicts? Then we'll take a look.

[OptimistOtaku]

Rebased onto main (6fbbaa577); GitHub now reports no conflicts. The resolution retains the marker exports added on main and this PR's cursor tests. Formatting, cargo check, 286 library tests, four integration tests and all-target Clippy pass locally. I also fixed an inherited shape-test initializer that Clippy flagged after the rebase. Fresh CI is running.

@greptileai please review the rebased head, 1f12a8c5229a5f6b9377b4d461d8588801106563, so the review reflects the current commit.