Analysis issues for cyclone

Abstract

Cyclone [1, 2] is an experimental, type-safe programming language based upon the syntax, semantics, and spirit of C. The primary goal of the language is to provide a type-safe environment that is close enough to C in both appearance and functionality, that systems programmers will find it attractive and useful.The most challenging aspect of the design is capturing the spirit of C without compromising type-safety. In particular, systems programmers expect to have good control over data representations, memory management, and performance. Yet, these features are usually absent from high-level, type-safe languages (e.g., Java). Another challenge is validating a sufficiently wide set of idioms that are in fact type-safe, but which conventional type systems reject.To address these issues, we have used a novel combination of typing features in conjunction with some interesting inference and dataflow techniques. The most novel typing feature is the support for region-based memory management which was summarized in an earlier paper [1]. However, this paper did not discuss the inference techniques we use to validate the regions and effects.In this talk, I will briefly summarize the Cyclone type system and then focus on the analysis issues that arise in its implementation, including (a) kind and type inference, (b) region and effect inference, and (c) dataflow analysis for validating initialization, array subscripts, and linear pointers.