Detecting fragile comments

Abstract

The development lifecycle of a software system demands incessant improvements in the source code of a system to maintain its high quality with improved performance and code readability.Refactoring is a common software development practice that reshapes the internal structure and non-functional properties of a system without modifying its core functionality. Many simple refactorings like renaming code elements, extracting a snippet from large method to form new method etc. can be performed with the help of automatic tools. Renaming code elements like classes, interfaces or methods is a widely used refactoring activity. With tool support, rename refactorings can rely on the program structure to ensure correctness of the code transformation. Unfortunately, the textual references to the renamed identifier present in unstructured comment text cannot be formally detected through the syntax of the programming language. These textual references to the previous version of a renamed identifier pose threats to the consistency between code and comments, which leads to poor program comprehensibility. The comments containing such textual references become fragile with respect to the renamed program element and are referred to as fragile comments.This thesis proposes a new rule-based approach to detect and fix the fragile comments that result from renaming the identifiers. We implemented this approach for the Java programming language in the form of an Eclipse plug-in called Fraco. Fraco takes into account the type of an identifier, its morphology i.e. the part-of-speech tag and its inflectional form, its scope that defines its visibility in the source code and the location of comments in the source code with respect to the identifier.i We evaluated the performance of our technique, as implemented for Java in Fraco, by comparing its precision and recall against hand-annotated benchmarks created for both development and test sets each containing six target Java systems, and also compared the results against the performance of Eclipse's automated in-comment identifier replacement feature. Fraco performed with an average of 99% precision and recall on most components of both development and test data sets, and generally outperformed the baseline Eclipse feature. An average percentage of 25% of the total identifiers of category type and method in the data sets had fragile comments after renaming, which further motivates the need for research on automatic comment refactoring. ACKNOWLEDGMENTSThe completion of this thesis has been a great learning journey which has left me indebted to many people who have supported, contributed and guided me, many a times, by going out of their way in doing so.First and foremost, I would like to thank my supervisor Prof. Martin P. Robillard for his unconditional support, methodical guidance and utmost care towards the ideation, development and completion of this thesis. Being confidently able to finish this thesis, I respect and admire his patience in successfully training and repeatedly enc...

What do class comments tell us? An investigation of comment evolution and practices in Pharo Smalltalk

“…Given the relevance of code comments for program comprehension and maintenance activities (Woodfield et al 1981;Tenny 1985;Tenny 1988;Hartzman and Austin 1993;de Souza et al 2006;Lidwell et al 2010;Cornelissen et al 2009), researchers have analyzed comments to detect low-quality comments (Steidl et al 2013;Liu et al 2015), identify existing inconsistency between comments and their related code elements (Ratol and Robillard 2017;Wen et al 2019;Stylos et al 2009;Petrosyan et al 2015;Zhou et al 2017), and they have examined the co-evolution of comments and code (Jiang and Hassan 2006;Fluri et al 2007;Fluri et al 2009;Ibrahim et al , 2012). However, very few studies have focused on analyzing the information embedded in the source code comments (Padioleau et al 2009;Haouari et al 2011;Steidl et al 2013;Pascarella and Bacchelli 2017;Zhang et al 2018), and none of them specifically analyzed class comments, or to what extent these class commenting practices adhere to the coding style guidelines.…”

Characterising the Knowledge about Primitive Variables in Java Code Comments

“…Preprocessing variables' identifiers and comments is an essential step for our detection approaches. For this, we followed part of the steps proposed by Ratol et al [17] for preprocessing both variables' identifiers and the text of comments, while we create our own prepossessing steps to suit our matching techniques. We present these steps as following:…”

“…Six project repositories were used to evaluate our proposed approaches, motivated by previous work of Ratol et al [17] to detect fragile comments. These projects (see Table III) were selected because they were used in Ratol et al's work [17] and showed several advantages such as their availability as open-source, a diversity of application domains, and being well-commented. To allow further investigation, the six projects were cloned locally to ease the process to analysing them.…”

“…To allow further investigation, the six projects were cloned locally to ease the process to analysing them. The same work by Ratol et al [17] was followed to determine our sample size (i.e., we utilized the stratified random sampling strategy to achieve diversity of the variables in our sample). Stratified random sampling can prevent bias while ensuring that all classes of interest are covered in a sampled population.…”

“…A rule-based approach named Fraco was proposed by Ratol et al [17] to detect code-comment inconsistencies resulting from rename refactoring operations performed on identifiers. Their evaluation shows the superior performance achieved by Fraco as compared to the rename refactoring support implemented in Eclipse.…”