JuliaCon 2026

Specialization behavior in Julia can be surprising and mystifying at times, especially when it leads to inference failures or performance problems. The compiler has a number of techniques and heuristics that it uses to determine how code ought to specialized to maintain performance and code size.

This talk will focus on how the Julia compiler makes specialization decisions and where these can sometimes go wrong, leading to suboptimal code, type-unstable code, --trim failures, and more.

To illustrate the discussion, we'll see scenarios where specialization goes right (as well as where it "goes wrong") and we'll dive into what it looks like to try to guide the compiler to specialize the way you need / want it to.