Source code patches from dynamic analysis

Abstract

Dynamic analysis can identify improvements to programs that cannot feasibly be identified by static analysis; concurrency improvements are a motivating example. However, mapping these dynamic-analysis-based improvements back to patch-like sourcecode changes is non-trivial. We describe a system, Scopda, for generating source-code patches for improvements identified by execution-trace-based dynamic analysis. Scopda uses a graph-based static program representation (abstract program graph, APG), containing inter-procedural control flow and local data flow information, to analyse and transform static source-code. We demonstrate Scopda's ability to generate sensible source code patches for Java programs, though it is fundamentally language agnostic.

Refactoring traces to identify concurrency improvements

“…In the future we plan to develop more optimisations, methods for addressing the combinatorial explosion of improvements, and support for concurrency models. A companion paper describes Scopda [9] which considers the problem of generating git-style source code patches for the optimisations suggested by Rehype.…”

“…A subtlety is that Rehype does not detect coordinator tasks, rather it estimates the effect of converting each applicable task (those that create and wait for another "sub-task") to a combinator. This embodies the "identify improvements by estimating the performance effects of implementing them" key idea described in the introduction; we leave determining and proposing concrete patches to the companion tool Scopda [9].…”

“…Rehype produces improvement specifications which define how to implement an improvement; it does not generate source code patches. A companion paper [9], also appearing at FTfJP'21, describes a system, Scopda, for generating source code patches from Rehype improvement specifications.…”

Augmenting Diffs With Runtime Information

“…Based on the collected data they output potential inefficient code in the program. They also introduce SCOPDA which uses REHYPE output and suggests concrete improvements to the program [37]. These two tools work with execution traces of Java programs, however their goal is different from COLLECTOR-SAHAB: COLLECTOR-SAHAB uses execution differences to explain code diffs, while SCOPDA analyzes execution trace to improve the efficiency of programs.…”