Measure draw time

Drawing writes straight into ratatui’s buffer: no allocation, no clock, no input and no randomness, so a frame costs microseconds and the same state always draws the same cells. A counting allocator in the crate’s tests holds that drawing and hit-testing allocate nothing.

Run the bench

From a clone of the repository, in release:

cargo run --release --example bench 200000

The argument is the number of frames per case (20,000 when left out). It draws five cases into a 40-row buffer and prints the mean and the worst frame of each:

pito-header bench: nav rows: 200000 frames at 150x40: mean 1.75 µs, worst 117.1 µs
pito-header bench: full header: 200000 frames at 150x40: mean 6.81 µs, worst 138.0 µs
pito-header bench: full header, drilled in: 200000 frames at 150x40: mean 4.39 µs, worst 1013.7 µs
pito-header bench: full header, abbreviated: 200000 frames at 40x40: mean 4.69 µs, worst 598.3 µs
pito-header bench: hit test: 200000 frames at 150x40: mean 0.13 µs, worst 77.4 µs

That run was on an AMD Ryzen 9 9950X. The worst frame is mostly the one the operating system interrupted and moves from run to run, so read the mean.

Case	What it draws
nav rows	the groups and sections rows alone, at 150 columns
full header	title with both slots, nav rows, a facts rule, a breadcrumb row and a notice on every other frame
full header, drilled in	the same, two levels deep, so the breadcrumb replaces the sections
full header, abbreviated	the full header at 40 columns, where every row abbreviates
hit test	one click hit-test on the nav rows, its column moving each frame

The chart above is the mean of 200,000 frames per case, measured on 2026-10-14 with Rust 1.97.1 and drawn by pito-compare.