Abstract

This talk is in two halves. In the first, I consider two problems with virtual machine-based language runtimes. Specifically, foreign function interfacing is needlessly difficult, and meanwhile, tools such as debuggers, profilers etc., being /per-VM/ infrastructure, create analogous difficulties for developers crossing VM—native or VM—VM boundaries. The focus is on dynamic languages; specifically, I will describe DwarfPython, an implementation of Python which aims to overcome both of these problems through some unusual design choices. Central among these is the adoption of a unifying descriptive metamodel (DWARF) from native debugging infrastructure. In the second half, I will extend these ideas to argue that “virtual” execution environments may be more accurately considered as performing “augmented execution”. From this observation, and by reconsidering the boundaries between dynamic and static analyses, I will suggest how to uncover apparently new spaces in the various trade-offs around both execution strategies and analysis techniques.

This talk (especially the second half) is very speculative; please bring your discussion hats!