Abstract

Using an optional type system requires significant manual effort in the form of writing type annotations. We have implemented a tool to automatically generate these annotations. Our approach observes and collects data about a running program, which we combine and summarize in the form of type annotations.

The naive approach to observing running programs is expensive. In this talk, we present some key optimizations analogous to space-efficient gradual typing that significantly reduces this expense, with minor tradeoffs in annotation accuracy.

Slides