An Empirical Study of Patch Uplift in Rapid Release Development Pipelines

Requirements

• Python 2.7 or newer
• Python 3.4 or newer
• R 3.2 or newer

Research questions

• RQ1: What are the characteristics of patches that are uplifted?
• RQ2: How effective are uplift operations?
• RQ3: What are the characteristics of uplifted patches that introduced faults in the system?
• RQ4: Are regressions caused by uplift more severe than the bugs that were fixed with the uplift?
• RQ5: Could some of the regressions have been prevented through more exten- sive testing on the channels?

Data mining scripts

• bug_inducing.py: identifies fault-inducing patches based on the SZZ algorithm.
• complexity_sna folder: compute source code metrics againt the Understand and igraph tools. The results will then be extracted and pretty outputted by src_code_metrics.py.
• analyze_bugs.py: extract some metrics about review and uplift process, developer/reviewer familiarity, etc..
• senti_metrics.py: extract sentiment metrics from the comments in the issue reports.
• review_metrics.py: extract code review-related metrics from comments and patch flags in the issue reports.

Data Analysis scripts

• comparison_acceptation.py: performs Mann-Whitney U test and Cliff's data effect size analysis to compare the characteristics between accepted pathces and rejected patches for uplift.
• comparison_failure.py: performs Mann-Whitney U test and Cliff's data effect size analysis to compare the characteristics between fault-inducing patches and clean patches that are uplifted.
• sample_analyses.ipynb: shows the statistics on the manual analysis results of RQ2, RQ4, and RQ5.

Data folder

• independent_metrics folder contains all metrics calculated for statistical analyses along various dimentions.

How to user the analytic scripts

1. Generate commit_date.csv by executing the following command: hg log --template '{node|short}\t{date|isodate}\n' > commit_date.csv.
2. Consecutively execute the data mining scripts to extract data for statistical analyses.
3. Execute the data analysis scripts to compare characteristics between accepted vs. rejected patches (RQ1), and between fault-inducing vs. clean patches (RQ2).

Data source

• Bugzilla: https://www.bugzilla.org
• Mozilla source code: https://developer.mozilla.org/en-US/docs/Mozilla/Developer_guide/Source_Code/Downloading_Source_Archives

Tool

• Understand: https://scitools.com/
• igraph: http://igraph.org

Other

• presentation_ICSME.pdf presentation slides at ICSME 2017.