Gradual Typing in an Open World

Gradual typing combines static and dynamic typing in the same language, offering the benefits of both to programmers. Static typing provides error detection and strong guarantees while dynamic typing enables rapid prototyping and flexible programming idioms. For programmers to take full advantage of a gradual type system, however, they must be able to trust their type annotations, and so runtime checks must be performed at boundaries between static and dynamic code to ensure that static types are respected. Higher order and mutable values cannot be completely checked at these boundaries, and so additional checks must be performed at their use sites. Traditionally, this has been achieved by installing proxies on these values to moderate the flow of data between static and dynamic code, but this can cause problems if the language supports comparison of object identity or foreign function interfaces. Reticulated Python is a gradually typed variant of Python 3 implemented via a source-tosource translator. It implements a proxy-free design named transient casts. This paper presents a formal semantics for transient casts and shows that not only are they sound, but they work in an open-world setting in which the Reticulated translator has only been applied to some of the program; the rest is untranslated Python. We formalize this open world soundness property and use Coq to prove that it holds for Anthill Python, a calculus that models Reticulated Python.