Abstract

Most software continues not to be soft. Most software, perhaps not coincidentally, is built atop Unix-like abstractions originating in the 1970s. To surmount this, instead of treating our Unix-like core as a black box (as language VMs do) or throwing it away (as the world has steadfastly refused to do) we can choose to evolve these abstractions, compatibly. Done right, this can give new 1980s-style superpowers (or things even more futuristic still!) to all software, not just new software built with new systems or languages. I’ll talk about some new support for “live” change I’m prototyping within a commodity Linux userspace: object motion via pointer tracking, source-in-situ (borrowing the best properties of Smalltalk images), and reflection down to the source level. Along the way, we’ll visit some of the murkier corners of the Unix userland, such as what goes on during dynamic linking, how debugging works and doesn’t (currently), and why linking and garbage collection amount to the same problem.

Bio

Stephen Kell does practical research on programming systems. His interests span most aspects of how we build software. and his research goal is to create infrastructure that makes quality software more cost-effective to create – where the costs in question are primarily human, not about machine resources. The systems he builds sit mostly on the boundary between operating systems and language runtimes, between which he sees only a very fluid boundary.

Some past and present research topics include: building software out of heterogeneous “found” components; robust dynamic analysis on the JVM; cross-language interoperability; realistic formal and metaprogrammable specifications of operating systems’ linking, loading and system call interfaces; reflective run-time services in Unix-like processes; and using the latter to provide new kinds of dynamic checking in C and other [traditionally] unsafe languages.

Currently he is a Lecturer at the University of Kent.