• Sources: arXiv, discussion
  • Summary: The author of the mold linker published a paper on its design, reporting speedups of 2.4x to 16.1x over lld. The paper is accepted to ASPLOS 2027. The ablation reports that no single optimization dominates the result.
  • Why it matters: The speedup comes from parallelizing every pass rather than from one trick another linker could adopt, which sets expectations for what incremental work on existing linkers can recover.