Abstract

I will discuss three ways in which program analysis can be considered ‘personal’. The first is to present every author of a software system with a report of the types of problems they tend to introduce and the types they tend to fix, throughout the entire history of a software project: experience shows that developers are much more willing to fix issues that they introduced themselves. The second is to allow a programmer without interpreter/compiler experience to begin writing their own analyses quickly (in the form of concise queries written in QL, a modern Datalog variant), then iteratively improving them by trying them on many non-trivial software projects simultaneously: every code base is special, and the only way to achieve high precision is extensive experimentation. The third is to provide an easy way of customising highly complex analyses to the peculiarities of a particular code base: once again, every code base is special, and one size does not fit all.

All this will be demonstrated using a service called CodingStars, developed by Semmle Ltd in Oxford (currently in private beta), along with several technical challenges and their solutions.