Integrating typed and untyped code in a scripting language

Abstract

Many large software systems originate from untyped scripting language code. While good for initial development, the lack of static type annotations can impact code-quality and performance in the long run. We present an approach for integrating untyped code and typed code in the same system to allow an initial prototype to smoothly evolve into an efficient and robust program. We introduce like types, a novel intermediate point between dynamic and static typing. Occurrences of like types variables are checked statically within their scope but, as they may be bound to dynamic values, their usage is checked dynamically. Thus like types provide some of the benefits of static typing without decreasing the expressiveness of the language. We provide a formal account of like types in a core object calculus and evaluate their applicability in the context of a new scripting language.

The ins and outs of gradual type inference

“…There has been a lot of interest in exploring ways to mix typed and untyped code, e.g., via occurrence typing [22], gradual typing [16,17], hybrid typing [8], and like typing [26]. In these systems, types are supplied by the user.…”

Tackling the efficiency problem of gradual typing

“…Theoretical approaches have been developed to tackle the space dimension [7,13], but execution time is also an issue. This has led certain languages to favor a coarse-grained integration of typed and untyped code [15] or to consider a weaker form of integration that avoids costly casts [16]. * Esteban Allende is funded by a CONICYT-Chile Ph.D.…”

Always-available static and dynamic feedback

“…Thorn's "like types" [7,47] are intermediate between staticallychecked types and Dynamic, and their benefits are also intermediate between the two. A like type's uses must be correct according to the static type system, but any value may be assigned to a like type (and must be checked at runtime).…”

“…Some of the related work assumes that a programmer will want to use dynamic typing indefinitely in some parts of a program; the work seeks interoperability with staticallytyped code. Other work aims to evolve dynamically-typed programs into statically-typed programs [5,9,39,44,45,41,7,47]. But, once the program has been converted, the benefits of dynamic typing are lost during later maintenance stages (unless the programmer explicitly changes types back to Dynamic).…”

“…Human effort and attention dominate the cost of most software produced today, especially if one includes the costs of bugs and rework that result from inadequate or poorly-focused human attention. Much other work is motivated by efficiency, even to the point of removing features like blame assignment [47]. By contrast, our primary focus is on the programmer.…”