Skip to content- Sources: primary, discussion
- Summary: A rustc patch series behind
-Zearly-metadata splits analysis into item interfaces and function bodies, writes an .early-rmeta file between the two, and swaps in full metadata before code generation, while a cargo series behind -Zheadstart starts dependents on the early-metadata notification and returns a paused compilation's job slot. The author measures clean builds of 13 real projects including rust-analyzer, zed, bevy, lemmy, and polars at up to 54 percent faster for cargo check and up to 42 percent for cargo build on 16 cores with none slower, falling to 24 percent for rust-analyzer's check on 4 cores because the gain comes from cores the build would otherwise leave idle, against stated costs of discarded downstream work, errors reported slightly later, and higher peak memory. Treat it as unlanded: the patches are described as a commit series meant to become upstream pull requests, the repository has 23 stars, and every figure is the author's own, though the repository ships scripts that compare diagnostics, JSON output, and exit status with the feature off and on and sweep the 53 rustc-perf compile benchmarks. - Why it matters: Rust clean-build time is a standing complaint, and this is a concrete scheduling mechanism rather than a tuning tip.
- Follow-up: Watch whether the early-metadata and headstart series open as upstream rustc and cargo pull requests.