Detection of software modules with high debug code churn in a very large legacy system

Abstract

Socaety has become so dependent on relaable telecommunacataons, that faalures can rask loss of emergency servace, busaness dasruptaons, or asolataon from fraends. Consequently, telecommunacataons software as requared t o have hagh relaabalaty. Many prevaous studaes define the classaficataon fault-prone an terms of fault counts. Thas study defines fault-prone as exceedang a threshold of debug code churn, defined as the number of lanes added or changed due t o bug fixes. Prevaous studaes have characterazed reuse hastory wath sample categoraes. Thas study quantafied new functaonalaty wath lanes of code. Thas paper analyzes two consecutave releases of a large legacy software system f o r telecommunacataons. W e applied dascrimanant analysas t o adentafy fault-prone modules based on 16 statac software product metracs and the amount of code changed durang development. Modules f r o m one release were used as a fit data set and modules f r o m the subsequent release were used as a test data set. I n contrast, comparable praor studaes of legacy systems splat the data t o szmulate two releases. W e valadated the model wath a realastac samulataon of utalazataon of the fitted model wath the test data set. Model results could be used t o gave extra attentzon t o fault-prone modules and thus, reduce the rask of unexpected problems.

Evaluating defect prediction approaches: a benchmark and an extensive comparison

“…Khoshgoftaar et al (1996) classified modules as defect-prone based on the number of past modifications to the source files composing the module. They showed that the number of lines added or removed in the past is a good predictor for future defects at the module level.…”

Comparing fine-grained source code changes and code churn for bug prediction

“…SCC, as used in this study, falls under this category. Among the first to study the relation between code churn defined as LM and Bugs was [17]. Work carried out in [25] explored the extent to which the use of relative code churn measures, e.g., LM weighted by total lines of code, outperformed absolute measures when predicting defect density: In Windows Server 2003, absolute churn measures showed a lower performance compared to relative ones.…”

A metric for software readability

“…We also note that machine learning has, in the past, been used for defect prediction, typically by training on data from source code repositories (e.g., [7,17,23,25]). We believe that machine learning has substantial advantages over traditional statistics and that much room yet exists for the exploitation of such techniques in the domains of Software Engineering and Programming Languages.…”