Efficient Selection of Type Annotations for Performance Improvement in Gradual Typing

Gradual typing allows mixing static and dynamic typing in a single language, but partially typed programs can suffer unexpected performance degradation when type annotations are added naively. This research presents a technique to selectively choose a subset of type annotations—derived via type inference—to improve execution performance in gradually typed programs. The approach uses a lightweight, amortized method that selects annotations along data flows to minimize expensive runtime casts at typed/untyped boundaries. Evaluated on Reticulated Python, the technique outperforms using all inferred annotations and achieves comparable execution speed to existing approaches while maintaining more stable compilation times.