Benchmarks

Manzil now ships a small benchmark harness for comparing the bundled linker against smfh on the same synthetic manifest.

Run it

nix run .#manzil-bench -- 2000 20

Arguments:

1. entry count
2. run count passed to hyperfine

Example:

nix run .#manzil-bench -- 5000 30

What it measures

The harness generates two manifests over a temporary tree and compares:

• activate
• diff

The synthetic workload mixes:

• symlink entries
• copy entries
• directory entries

Some entries stay unchanged between the old and new manifests, while others change source paths or permissions. That gives a useful approximation of a real activation instead of a single toy file.

Notes

• the benchmark uses temporary absolute paths, not /nix/store
• results are only meaningful when run on an otherwise idle machine
• the benchmark is for linker throughput, not Nix evaluation time
• the benchmark is intentionally reproducible enough for local iteration, not a formal performance lab

If you make linker changes, run the harness before and after the change and keep the exact command in the commit or PR notes.