A study of code change patterns for adaptive maintenance with AST analysis

Abstract

Example-based transformational approaches to automate adaptive maintenance changes plays an important role in software research. One primary concern of those approaches is that a set of good qualified real examples of adaptive changes previously made in the history must be identified, or otherwise the adoption of such approaches will be put in question. Unfortunately, there is rarely enough detail to clearly direct transformation rule developers to overcome the barrier of finding qualified examples for adaptive changes. This work explores the histories of several open source systems to study the repetitiveness of adaptive changes in software evolution, and hence recognizing the source code change patterns that are strongly related with the adaptive maintenance. We collected the adaptive commits from the history of numerous open source systems, then we obtained the repetitiveness frequencies of source code changes based on the analysis of Abstract Syntax Tree (AST) edit actions within an adaptive commit. Using the prevalence of the most common adaptive changes, we suggested a set of change patterns that seem correlated with adaptive maintenance. It is observed that 76.93% of the undertaken adaptive changes were represented by 12 AST code differences. Moreover, only 9 change patterns covered 64.69% to 76.58% of the total adaptive change hunks in the examined projects. The most common individual patterns are related to initializing objects and method calls changes. A correlation analysis on examined projects shows that they have very similar frequencies of the patterns correlated with adaptive changes. The observed repeated adaptive changes could be useful examples for the construction of transformation approaches

Do API-Migration Changes Introduce New Bugs?

“…Therefore, we have categorized retrieved fixing commits as adaptive or non-adaptive using the results of our previous studies (Meqdadi et. al., 2013;Meqdadi et. al., 2020).…”

“…Therefore, we reused the categorization of adaptive commits performed in (Meqdadi et. al., 2013;Meqdadi et. al., 2020) to mark adaptive commits that would be described as bug-introducing commits.…”

“…As well, we have previously inspected the undertaken adaptive maintenance changes of these selected systems in our prior research (Meqdadi et. al., 2013;Meqdadi et. al., 2020).…”

“…If a message of a commit encloses key terms of API-migration tasks (i.e., APIs interfaces or language features), then the commit is considered as an adaptive commit. Similarly, the adaptive commits of the remaining systems were allocated and examined in and (Meqdadi et. al., 2020), where the study proposed a machine learning classifier to label the commits of an evolution history into either one of two classes, namely adaptive and non-adaptive commits.…”

Changes from the Trenches: Should We Automate Them?

“…These changes can be unique; however, in many cases, they are repetitive and follow patterns [24]. Such patterns can be a rich source of information for analyzing the history of changes and their impact [10], modification records of fault fixes [7], or code change patterns' relationship with adaptive maintenance [18].…”