Evidence: Vite React dev-server benchmark

Every number in 4 Things Making Your Vite Dev Server Slow comes from these files. Measured on an Apple M5 (10 cores), Node v24.16.0, Vite 7.

Results

File	What's in it
bench-results.json	The five-configuration matrix: cold start (5 runs), first-load module-graph crawl, HMR round-trip, production build (3 runs), and dist chunk tables. Records machine specs and dependency versions.
sweep-results.json	Module-count sweep from 68 to 3,059 modules against both React plugins, 3 runs per cell, plus the fitted per-module cost and the real `@mui/icons-material` comparison.
mui-warm.json	Cold vs warm dependency-cache runs for the MUI icon barrel — the test that shows most of its cold cost is esbuild pre-bundling.

File	What it does
gen.mjs	Generates the fixture: N components behind a barrel `index.js`, 50 SVG icons, three app variants. Size set by `FIXTURE_COMPONENTS` (default 3000).
bench.mjs	The five-configuration matrix. Starts a real dev server per run, crawls the entry module graph the way a browser does, drives HMR over Vite's websocket.
sweep.mjs	The module-count sweep, the linear fit, and the third-party package comparison.
mui-warm.mjs	Cold/warm dependency-cache test for the pre-bundling hypothesis.
charts.mjs	Renders the article's SVG charts directly from the results JSON — no numbers are typed by hand.

The fixture source is generated, so it isn't committed. From evidence/articles/vite-react-performance/fixtures/ in the PageSpeedFix repo:

node gen.mjs
npm install
node bench.mjs
node sweep.mjs
node mui-warm.mjs

Absolute timings depend on your machine. The ratios, the linear fit, and the module counts should reproduce.