Roc rewrites its 300k line compiler from Rust to Zig

Sources: rtfeldman.com, HN discussion
Summary: Richard Feldman published an account on 2026-07-15 of scratch-rewriting the Roc language compiler, about 300,000 lines, from Rust to Zig, reaching feature parity after 487 days. The post cites Cargo build times that degraded as the codebase grew, fundamental bugs in lambda-set resolution that needed cross-phase rewrites regardless, and Zig's allocator model fitting the team's arena allocation and struct-of-arrays layouts. It reports incremental builds of 3.4 seconds on Rust 1.97.0 versus 35 milliseconds on a Zig 0.17.0 nightly, and counts 10 memory-corruption bugs (2 use-after-free, 8 miscompilations) among 431 Zig-compiler bugs against 21 memory-corruption bugs among 2,596 Rust-compiler bugs, arguing most of its memory-safety issues traced to miscompilation rather than unsafe compiler code.
Why it matters: It is a first-party rewrite report that runs opposite to Bun's Zig-to-Rust move the same month, and it separates the build-time and memory-safety outcomes that these language debates usually argue in the abstract.