Rui Hara's mold linker has shipped version 3.0.0, and the headline is the rewrite: the entire codebase has been ported from C++ to Rust. This is not an incremental refactor or a mixed-language experiment. The C++ version (2.42.1) is now end-of-life. The Rust version is intended as a drop-in replacement — same command-line options, same target architectures, same output. The project verified compatibility by running its test suite across every supported target, comparing output across real-world workloads, and building all Gentoo packages. No regressions were found. The strategic goal is explicit: close the remaining compatibility gaps with GNU ld, particularly in linker script support, and pave the way for mold to be adopted as the default linker in Linux distributions. This is a direct challenge to the GNU toolchain's decades-long incumbency in the critical-path build infrastructure of every major Linux distro. If mold succeeds, it changes the foundational plumbing of how Linux software is assembled. The Rust rewrite delivers a concrete safety improvement beyond language ideology. The C++ version could read memory out of bounds on corrupted input files and crash with a segfault. The Rust version bounds-checks those reads, panicking at the faulty access instead of silently producing undefined behavior. For a tool that processes untrusted object files from potentially adversarial build inputs, this is not cosmetic. The build system has changed from CMake to Cargo, requiring Rust 1.95+ and a C compiler. The oneTBB dependency is gone. mimalloc 3.5.3 is statically linked by default, with a system-allocator feature flag available. These are packaging decisions that matter for distro adoption — fewer external dependencies means fewer integration headaches for maintainers. The bugfix list is substantial and tells its own story about the project's maturity ambitions. Fixes span nondeterministic output, incorrect GOT-relative relocations, range extension thunk errors in large programs like Chromium's ARM64 debug build, and platform-specific issues across AArch64, ARM32, RISC-V, LoongArch, PPC32, PPC64, SH4, and SPARC64. Several previously silent wrong-output bugs are now reported as errors. The breadth of architecture coverage signals serious intent to be a universal linker, not just an x86-64 speed trick. The project remains open-source and funded through GitHub Sponsors and OpenCollective, with corporate sponsors including Cybozu, SAP, Mercedes-Benz, and Ahrefs. This is a lean sponsorship model for infrastructure software that sits in the critical path of every compiled binary on Linux. The question is whether this funding model can sustain the maintenance burden of being a distro default. Mold 3.0 is the rare infrastructure rewrite that ships with receipts: same output, full test coverage, no performance regression, and a clear roadmap. The hard part isn't the code — it's convincing distro maintainers that a Rust-based linker with a small team and sponsorship funding deserves the trust currently held by GNU ld and LLVM's lld.
