Turbopack: Building faster by building less

Summary

Edit. Save. Refresh. Wait… Wait… Wait… Compiling code usually means waiting, but Turbopack makes iteration loops fast with caching and incremental computation. Not every modern bundler uses an incremental approach, and that’s with good reason. Incremental computation can introduce significant complexity and opportunities for bugs. Caches require extra tracking and copies of data, adding both CPU and memory overhead. When applied poorly, caching can actually make performance worse. Despite all of this, we took on these challenges because we knew that an incremental architecture would be critical to Turbopack’s success. Turbopack is the new default bundler for Next.js, a framework that is used to build some of the largest web applications in the world. We needed to enable instant builds and a fast as-you-type interactive React Fast Refresh experience, even for the largest and most challenging workloads. Our incremental architecture is core to achieving this. A solution without caching slows down as the size of the application increases. Incremental computation can scale to large applications because it only depends on the size of small iterative changes. Turbopack’s architecture was built ground-up with caching in mind. Its incremental design is based on over a decade of research. We built on first-hand experience from challenges in implementing caching in webpack and drew inspiration from Salsa (which powers Rust-Analyzer and Ruff), Parcel, the Rust compiler’s query system, Adapton, and many others. Turbopack achieves a fine-grained cache by automatically tracking how internal functions are called and what values they depend on. When something changes we know how to recompute the results with minimal work. Background: Manual incremental computation Many build systems include explicit dependency graphs that must be manually populated when evaluating build rules. Explicitly declaring your dependency graph can theoretically give optimal results, but in practice it leave...