Efficient Manipulation of Control Flow Models in Evolving Software

Efficient Manipulation of Control Flow Models in Evolving Software

Paper in proceedings (conference paper)

When looking for certain kinds of software bugs, successive versions of software are compared. Performance-related bugs are a notable example. Methods used for detecting such bugs are, however, expensive and need to be applied carefully. At the same time, current software development is rapid, with new software versions released everyday. In this paper, we aim at two particular ways how to optimize difference analyses of performance (but possibly other aspects of the software too) of successive software versions. Namely, we propose (1) an efficient layered representation of the program control flow spanning across the program history, and (2) methods for efficient matching of pairs of corresponding functions in different software versions and for selecting those whose differential analysis should be performed. We have implemented our approach and performed experiments on two selected versions of the CPython project. The results indicate that our approach is a promising direction for improving the performance analysis of real world programs.

When looking for certain kinds of software bugs, successive versions of software are compared. Performance-related bugs are a notable example. Methods used for detecting such bugs are, however, expensive and need to be applied carefully. At the same time, current software development is rapid, with new software versions released everyday. In this paper, we aim at two particular ways how to optimize difference analyses of performance (but possibly other aspects of the software too) of successive software versions. Namely, we propose (1) an efficient layered representation of the program control flow spanning across the program history, and (2) methods for efficient matching of pairs of corresponding functions in different software versions and for selecting those whose differential analysis should be performed. We have implemented our approach and performed experiments on two selected versions of the CPython project. The results indicate that our approach is a promising direction for improving the performance analysis of real world programs.

Performance analysis, Cross-version performance degradation, Control flow, Perun,

Performance analysis, Cross-version performance degradation, Control flow, Perun,

FIEDOR, T.; PAVELA, J.; ROGALEWICZ, A.; VOJNAR, T.

2026

24.04.2025

Springer International Publishing

Las Palmas de Gran Canaria

978-3-031-82949-9

Eurocast 2024 -- Computer Aided Systems Theory: Extended Abstracts

Lecture Notes in Computer Science

Federal Republic of Germany

412

427

16

https://link.springer.com/chapter/10.1007/978-3-031-82949-9_37