Repository navigation
headstart seems to run 2x SLOWER than nightly? #7
Description
Building the headstart/test/smoke project on nightly: 4.9s
Benchmark 1: cargo clean && cargo build
Time (mean ± σ): 4.909 s ± 0.100 s [User: 4.790 s, System: 0.555 s]
Range (min … max): 4.768 s … 5.062 s 10 runsBuilding the headstart/test/smoke project using headstart: 10.7s
Benchmark 1: cargo clean && RUSTC=~/Projects/headstart/headstart/rustc/build/host/stage1/bin/rustc ~/Projects/headstart/headstart/cargo/target/release/cargo build -Zheadstart
Time (mean ± σ): 10.969 s ± 0.339 s [User: 11.008 s, System: 0.544 s]
Range (min … max): 10.622 s … 11.820 s 10 runs2x slower on headstart?
Environment
Built headstart today:
$ RUSTC=~/Projects/headstart/headstart/rustc/build/host/stage1/bin/rustc ~/Projects/headstart/headstart/cargo/target/release/cargo --version
cargo 1.101.0 (4f3fb2428 2026-09-30)And I'm comparing it against with nightly:
$ cargo --version
cargo 1.101.0-nightly (3960fa69b 2026-10-12)Hardware:
6-core Ryzen™ AI 5 340
32GB Ram
Activity
0atman commentedon Oct 16, 2026
It's also 2x slower at a real-world test of compiling
ripgrep... maybe my machine has too few cores to be useful?zmaril commentedon Oct 20, 2026
Hello @0atman ! Thank you for pointing this out, it took me a bit to figure out why you might be getting those numbers. I've updated the readme (and added a script) to address this: https://github.com/PowderworksCode/headstart#comparing-with-nightly
Nightly Rust is built with Pgo and BOLT enabled, which makes it much faster than the dev version. Previously, setup.sh for headstart just built the dev version without Pgo and BOLT. A new script, scripts/release.sh, now builds the Pgo and BOLT version and when I measured it, it showed the advertised improvements over Nightly Rust.
So, could I ask you to try this out and see how it performs? Thank you very much for filing an issue, I should have addressed this earlier!