Bflinks: Reliable Bugfix Links via Bidirectional References and Tuned Heuristics.

Area of Science:

• Software Engineering
• Empirical Software Engineering
• Software Maintenance

Background:

• Software version archives and defect databases are crucial for analyzing defect causes and predicting future issues.
• Identifying bugfix commits and linking them to defect database entries is a foundational step for these analyses.
• Current methods often rely on heuristic string matching of commit messages, which can be imprecise.

Purpose of the Study:

• To identify effective filters for creating bugfix links between software version archives and defect databases.
• To determine optimal parameter tuning strategies for these filters.
• To evaluate the accuracy and efficiency of the developed bugfix linking approach.

Main Methods:

• Analysis of seven independent filters, including novel filters utilizing reverse links.
• Evaluation of visual heuristics for setting filter cutoff parameters.
• Manual verification of over 2500 links by a product expert on a commercial software repository.

Main Results:

• Heuristics successfully identified optimal or near-optimal parameter values for most filters.
• The combined filtering approach, termed bflinks, demonstrated high precision (93%) with minimal results loss (7%).
• The method proved effective in a real-world commercial repository setting.

Conclusions:

• Bflinks offers a robust and accurate method for establishing high-quality bugfix links.
• The approach is adaptable to the specific characteristics of different software repositories.
• This facilitates more reliable defect insertion circumstance analysis and defect prediction.