// the find
rui314/mold
mold 🦠: A Modern Linker in Rust 🦀
mold is a from-scratch linker (now written in Rust, after starting in C++) written by the original lld author, built to replace ld/lld/gold as a drop-in swap via -fuse-ld=mold. It's for anyone doing compiled-language builds (C/C++/Rust) where link time is the bottleneck in the edit-compile-link loop — the bigger your binaries, the bigger the win.
The benchmarks aren't cherry-picked toy cases — they link actual Chromium, Firefox, TensorFlow, and LLVM builds and show 5-16x over lld on a 64-core box, with real numbers for both debug and release configs on two different CPU architectures. Architecture coverage is unusually wide for a linker this young: x86-64, ARM32/64, RISC-V, PPC, s390x, LoongArch, SPARC64, m68k, SH-4, plus both ELF and Mach-O backends. It's not a research toy — production use since 2021, 140+ contributors, CI running the full test suite under ASAN/TSAN per target arch, and a pre-release pass building ~19,000 Gentoo packages against it to catch regressions before they ship. The `-run`/LD_PRELOAD trick is a genuinely useful escape hatch for build systems where you can't easily inject a linker flag.
No native Windows/PE-COFF target — if your toolchain is MSVC you get nothing here, only ELF and Mach-O. The speed advantage compresses hard on smaller release builds and on the M1 numbers specifically (several rows show mold within 10-20% of wild, one row where wild is actually faster), so the headline multipliers are mostly a debug-build-of-a-giant-binary story. GCC before 12.1 doesn't accept `-fuse-ld=mold` at all, forcing the awkward `-B/usr/local/libexec/mold` workaround — a real papercut if you're stuck on an older distro's toolchain. The project's own README is upfront that ongoing maintenance (new arch support, toolchain churn) depends on sponsorship money, which is worth knowing before you build critical CI infra around it.