benchmarks
phpm real-world benchmarks
Loading results…
Composer and phpm install the same real-world open-source PHP applications — the
corpus is bench-real/corpus.json,
40 projects by GitHub stars — cold, warm and no-op, 10 warm runs each, via
hyperfine, on GitHub-hosted
ubuntu-latest, macos-latest and windows-latest runners.
Every row also gets a diffvendor identity check and phpm's own
--explain fallback reasons. Losses are published: a project where phpm
falls back or is slower stays in the table with the reason.
Headline, by OS
Cold, warm, no-op
One horizontal bar chart per scenario: every project's Composer time (grey) against phpm's (amber), sorted by speed-up, ubuntu-latest.
Ten worktrees of laravel-skeleton
The AI-agent-workflow number: ten git worktrees of the same project, composer install × 10 (ten full vendor/ copies) against phpm install × 10 from one warm store.
| OS | Worktrees | Composer total | phpm total | Time speed-up | Disk, Composer : phpm |
|---|
CI job time, with and without actions/cache
A representative project's own CI install step, before and after phpm: total job time (install plus a lint pass over vendor/), not just the install line.
| OS | Project | Composer, cached | phpm, cached | Speed-up, cached | Speed-up, uncached |
|---|
Projects
| Project | Stars | Packages | OS | Cold | Warm | No-op | Identity | Fallback |
|---|