Mold: A Massively Parallel Linker

Let's be real: linkers are the ugly stepchild of the toolchain. Compilers get conference keynotes, new languages get hype cycles, but the poor bastard that stitches all those object files together is treated like plumbing—until it's slow, and then it's everyone's water main bursting at 4:55pm on a Friday. For decades, we just accepted that linking was a fundamentally serial, memory-bound horror show. Mold says nah, and the arxiv write-up is a refreshingly blunt post-mortem of why everybody else was wrong. The technical pitch is simple in hindsight: stop using a single global symbol table guarded by a giant lock, design for parallel insertion, use custom hash tables that don't suck, and actually exploit the fact that modern machines have 16, 32, 64 cores doing absolutely nothing while you sit there staring at a progress bar. The paper documents this the way a good systems paper should—no fluff, just "here's the bottleneck, here's how we dismantled it." It's the kind of work that looks obvious after the fact, which is exactly how you know it was genuinely hard. But let's talk about the drama, because that's what draws blood in this community. Mold isn't just a research curiosity—it's an AGPL/commercial dual-license landmine dropped into the middle of the GNU/Gold/LLD status quo. That kicked off actual fear in toolchain circles: suddenly the "free" linker everyone depended on had a fast competitor with a license that makes corporate lawyers reach for their paper bags. The HN thread (https://news.ycombinator.com/item?id=494