On the Time Performance of Automated Fixes

Abstract

Automated program repair has made major strides showing its exciting potential, but all efforts to turn the techniques into practical tools usable by software developers hit a crucial blocking factor: the timing issue. Today’s techniques are slow. Too slow by an order of magnitude at least. The long response time implies that the currently available techniques cannot suit the actual needs of developers in the field. What developers want is a tool that can instantaneously propose a fix for a detected failure. A technique that can propose a patch instantaneously (or near-instantaneously) would provide the breakthrough that is required to turn automated program repair from an attractive research topic into a practical software engineering tool. Indeed, researchers have started to tackle this speed issue. In this paper, we survey recent approaches that were shown to be effective in speeding up automated program repair. In particular, we view automated program repair as a search problem—the ultimate goal of automated program repair is the search for a patch which is often preceded by other related searches such as a search for suspicious program locations, and a search for the specification for a patch. We describe how the problem of automated program repair has been decomposed into a series of search problems, and explain how these individual search problems have been solved. We expect that our paper would provide insight into how to speed up automated program repair by further optimizing the search for a patch.