A developer has built a tail-call interpreter in nightly Rust that outperforms both previous Rust and hand-coded ARM64 assembly implementations. The project, part of an ongoing exploration of high-performance emulation of the Uxn CPU, leverages the newly stabilized `become` keyword to achieve performance gains previously only possible through assembly programming.

Uxn is a simple stack machine with 256 instructions and just over 64K of space split between two 256-byte stacks, 65536 bytes of RAM, and device memory. The developer's journey began with a basic Rust implementation, progressed through ARM64 assembly using threaded code techniques, and now culminates in this Rust solution. The assembly approach achieved 40-50% speedups on ARM64 and about 2× faster on x86-64, but required maintaining roughly 2000 lines of unsafe code.

The key breakthrough came from understanding tail-call optimization in Rust. By restructuring the interpreter to store program state in function arguments and using the `become` keyword, the compiler replaces stack frames instead of adding to them. This eliminates stack overflow issues while maintaining the performance benefits of threaded code. The developer notes that all tail-call code is human-written, despite previous experiments with LLMs proving controversial.