Type safety analysis for Dart

Abstract

Optional typing is traditionally viewed as a compromise between static and dynamic type checking, where code without type annotations is not checked until runtime. We demonstrate that optional type annotations in Dart programs can be integrated into a flow analysis to provide static type safety guarantees both for annotated and non-annotated parts of the code. We explore two approaches: one that uses type annotations for filtering, and one that uses them as specifications. What makes this particularly challenging for Dart is that its type system is unsound even for fully annotated code. Experimental results show that the technique is remarkably effective, even without context sensitivity: 99.3% of all property lookup operations are reported type safe in a collection of benchmark programs.

What is decidable about gradual types?

“…We have two categories of type inference: static and dynamic. Static type inference includes the inference tool for Python by Hassan et al [2018], and the inference tool for Dart by Heinze et al [2016]. Some approaches to type inference add types for the purpose of program understanding but without a guarantee that the resulting program type checks.…”

Fast and precise type checking for JavaScript

“…But we have enough anecdotal evidence from developers at Facebook that focusing on soundness is not only useful but also desirable, and does not necessarily imply inconvenience. Similar to Safe TypeScript, recent work recovers soundness in Dart [Heinze et al 2016]. Closure [2009] is another widely used type system for JavaScript that focuses on transforming code for size reduction.…”