Guides
Reproduce the benchmarks
Run the same harness that produces the numbers on this site, on your own machine.
Every performance figure on the site comes from bench/results/latest.json, produced by a harness in the repository. Run it yourself to check.
brew install hyperfine
swift build -c release
make bench-quick # S3 + S5, the headline scenariosmake bench runs all six scenarios. Fixtures pin a repository and commit each, and the harness runs in a temporary directory with its own store, so it never touches your ~/.spmx.
| Scenario | What it compares |
|---|---|
| S1 | cold resolve; spmx must not be slower |
| S2 | new worktree, SwiftPM cache warm |
| S3 | new worktree, store warm: native resolve vs spmx hydrate |
| S4 | physical disk per extra worktree |
| S5 | shim overhead on a no-op command |
| S6 | clean build with and without the shared compilation cache |
Disk is measured as the drop in free space of an isolated sparse APFS volume, not with du, which overstates clones.
Read the details, flags and caveats in bench/README.md, and see the results with their methodology on the benchmarks page.
To show your own run on a local copy of the site, copy bench/results/latest.json into place and rebuild:
cd website
npm run build # prebuild validates the JSON and fails if it is invalid