Abstract
C remains a very widely used systems programming language, with many advantages: widely known, supported by many tools, good access to low-level hardware, etc. But C also makes it unnecessarily hard to produce safe and reliable programs, a particularly significant problem for systems and multi-threaded code. Modern languages address many of these issues, but porting existing code to a new language is often impractical for large systems. Static analyses of existing C code can find some problems, but guaranteeing safety is hampered by extensive use of unsafe features.

In the Ivy project, we have designed small extensions to C that address the classic problems of type and memory safety (Deputy and HeapSafe), and the increasingly-important problem of data-sharing in threaded programs (SharC). All of these extensions require small, tractable changes to existing code, and have been validated on substantial code bases - a bootable Linux kernel for Deputy and HeapSafe, and several Linux-based threaded applications for SharC.

In this talk, I will give a brief overview of Deputy (type safety) and HeapSafe (deallocation safety), and concentrate on SharC (short for "Sharing Checker") that allows a user to write lightweight annotations to declare how they believe objects are being shared between threads in their program. SharC then uses a combination of static and dynamic analyses to check that the program conforms to this specification.