Java has supported generics since 2004, letting you write reusable algorithms and data structures that work across many types. A single List or Arrays.sort can handle strings, ints, or even frogs (if they’re comparable).

Generics may look like C++ templates, but they behave differently. Java must support both statically known and dynamically typed usage, meaning a single generic method balances flexibility with performance. How does the JVM optimize this? That’s the core question of this talk.

Using a QuickSort example, we explore how the JVM achieves high performance through type profiling, inlining, and devirtualization—often matching C++ template speed. But there are limits. Issues like profile pollution can degrade performance, causing code to behave more like dynamically typed execution.

We then look at techniques to avoid these performance cliffs and maintain efficiency.

This topic is especially relevant today with Project Valhalla. Valhalla introduces value classes, changing how data is laid out in memory. Arrays may no longer be simple pointer collections—they can resemble C++-style structs, making generics harder to optimize.

So how do we handle this? Short-term solutions exist, and we demonstrate them on today’s JVM (Java 25). Long-term, Valhalla aims to introduce generic specialization, allowing the JVM to generate optimized, type-specific versions of generic code—bringing Java closer to C++ template performance.

Join us as we explore generics, JVM internals, and the road to Valhalla generics.

Make sure to check the JavaOne 2026 playlist.