LLVM

LLVM (formerly the Low Level Virtual Machine) is a compiler infrastructure project that began at the University of Illinois in 2000 and has since become the dominant open-source compiler framework, powering the Clang C/C++ compiler, the Rust compiler, and the Swift compiler. Unlike traditional compilers that are monolithic black boxes, LLVM is a modular system of reusable libraries centered on a common intermediate representation called LLVM IR — a typed, portable assembly language that serves as the interchange format between compiler front ends and back ends.

The architectural bet of LLVM was that compiler research and production could share the same infrastructure. Before LLVM, academic compiler research typically built toy compilers that could not scale to real languages, while industrial compilers were closed-source systems that could not be modified by researchers. LLVM bridged this gap by providing a production-quality, modular optimizer that could be extended, replaced, and experimented upon. The result is that many modern compiler innovations — from link-time optimization to just-in-time compilation — were first prototyped in LLVM and then deployed in production.

Yet LLVM's dominance has created its own problems. The LLVM IR is not a neutral standard but a de facto monopoly, and its design choices — particularly its memory model and its type system — constrain what optimizations can be expressed. A compiler infrastructure that began as an alternative to monoculture has become a monoculture itself.