Rename Chains: An Exploratory Study on the Occurrence and Characteristics of Identifiers Undergoing Multiple Renamings
0 likes315 views
This study investigates the phenomenon of identifiers undergoing multiple renamings, known as rename chains, across 800 open-source Java systems. The research reveals that while most identifiers typically undergo a single rename, rename chains are common and primarily driven by code cleanup, refactoring, and bug fixes, often conducted by the same developer. Key findings include the median renaming interval, the grammatical structure of names in the chain, and the importance of high-quality names for code comprehension.
Rename Chains: An Exploratory Study on the Occurrence and Characteristics of Identifiers Undergoing Multiple Renamings
- 1. Rename Chains: An Exploratory Study on the Occurrence and Characteristics of Identifiers Undergoing Multiple Renamings Anthony Peruma & Christian Newman
- 2. Overview We explore the phenomenon of a single identifier undergoing multiple renames (i.e., a rename chain) through a large-scale empirical study of 800 open-source Java systems
- 3. Introduction • Research shows that developers spend 58% of their time on program comprehension activities • Identifier names account for 70% of the characters in the code base • Identifier names help developers understand the purpose of the identifier – essential that names should be high-quality • Names must be unambiguous and intent reveling in communicating the purpose and behavior of the code • Developers correct poor-quality names via rename refactoring operations – over 40% of refactoring operations are renames • Not all renames result in a high-quality name • An identifier can undergo multiple renaming's throughout its lifetime (i.e., a chain of renaming's – a rename chain)
- 4. Rename Chain Examples A method rename chain resulting in a more descriptive method name in the final rename A rename chain resulting in a weak method name; it is just a copy of the statement within the method
- 5. Related Work on Identifier Renaming • Empirical Studies • Arnoudova et al. – Rename taxonomy to classify the semantics updates to a name; developer study showing renaming is not always straightforward • Peruma et al. – Multiple studies that examine the structure and meaning of names → developers frequently narrow the meaning of the name; grammar patterns; contextualization; taxonomy for digits in a name • Recommendation Models • Allamanis et al. – utilizes statistical NLP to learn the coding style of a codebase • Suzuki et al. – An n-gram-based approach for assessing the comprehensibility of method names and recommending intelligible method names • Liu et al. – Deep learning techniques to provide recommendations based on the overlap between method bodies and names that are close in a vector space • … While there are studies that investigate (or involve) rename refactoring’s, this is the first study that examines a chain of rename operations for an identifier
- 6. Goal & Impact Understand the evolution of identifier names by constructing and studying the characteristics of a chain of renames for identifiers (i.e., a rename chain) Facilitate the research and development of tools to aid in name appraisal and recommendations
- 7. Research Questions • RQ1: To what extent do identifiers undergo multiple rename refactoring operations? • Understand the volume and types of identifiers that undergo multiple renames • RQ2: How frequently do renames occur within a rename chain, and who is responsible for their creation? • Gain insight into the developers performing the renames in the chain • RQ3: How do the semantics of an identifier's name evolve in a rename chain? • Determine the lexical-semantic properties of names in the rename chain • RQ4: To what extent can commit log messages help contextualize the occurrence of rename chains? • Identify the specific causes for developers to create rename chains
- 8. Contributions A publicly available dataset of rename chains for replication or extension studies An understanding of identifier name evolution and a discussion on avenues for future research
- 9. Experiment Design Source dataset of rename refactorings and commit details Dataset of rename chains and their characteristics Source Dataset: Used in prior refactoring research studies; renames mined using RefactoringMiner Rename Chain Construction: For each project: sort renames using the author commit date; compare the identifier’s old and new name by their fully qualified name Part-of-Speech Tagging: Utilize a specialized identifier name part-of-speech tagger; tags for only the original and last name in the rename chain Topic Modeling: Commit messages associated with rename chains; preprocessing; Latent Dirichlet Allocation RQ Analysis: Supplement our quantitative findings with qualitative examples from our dataset
- 10. Results
- 11. RQ 1: To what extent do identifiers undergo multiple rename refactoring operations? • Identifier renaming is a common operation developers perform • 285,786 operations: Methods: 26.50%; Parameters: 25.53%; Variables: 21.75% • Most identifiers undergo a single rename in their lifetime • 17,404 detected rename chains – Methods are likely to have a chain • Methods: 30.73%; Variables: 23.47%; Classes: 16.85% • A rename chain is usually short – composed of a median of 2 rename instances; variables typically undergo around 3 renamings • Rename chains tend to occur in projects frequently • 83.71% of projects have rename chains • Projects have a median of 9 identifiers undergoing multiple renames
- 12. RQ 2: How frequently do renames occur within a rename chain, and who is responsible for their creation? Interval Analysis – duration between renames in the chain • Median duration between renames is 2 days • Attributes: 25 days; Classes: 19 days; Methods: 14 days; Variables: 7 days; Parameters: 2 days • Duration between the first and last rename: • Parameters have the lowest interval: 17 days • Variables have the highest interval: 357 days Developer Analysis – developers performing the renames • Most chains have the same developer performing all renames: 62.05% • Multi-developer chains have around 2 developers involved • Attribute chains have the most number of developers: 4 • 11.51% of chains have different developers for the first and last rename
- 13. RQ 3: How do the semantics of an identifier's name evolve in a rename chain? Analysis of the lexical-semantic structure (i.e., part-of-speech tags) of names in the rename chain Analysis is limited to only the first and last names in the chain • The same part-of-speech tags are used for the original and new name • TestServlet → TheTestServlet → TestServlet : NM-N → DT-NM-N → NM-N • Developers utilize standard naming structures: • Class: NM-NM-N → NM-NM-N • Attribute: NM-N → NM-N • Method: V-NM-N → V-NM-N • Parameter & Variable: N → N • Usually, the original and new names are not identical – 78%
- 14. RQ 4: To what extent can commit log messages help contextualize the occurrence of rename chains? An automated analysis of rename chain commit log messages using Latent Dirichlet Allocation 3 high-level topics associated with these messages: • Code Cleanup – developer improving code style quality by adhering to standards – ‘naming’, ‘convention’, ‘whitespace’ • “Lots of fixes using Checkstyle - Fixed some names to follow conventions...” • Refactoring – developers updating the code related to the behavior and design of the system – ‘refactor’, ‘updated’, ‘revert’ • “Major refactor to start process of eventually moving content manager classes…” • Bug Fix/Testing – renames are part of either a bug fix or unit testing – ‘fix’, ‘bug’, ‘test’, ‘testcase’ • “fixed bug with searching for transitive dependencies + added test for it…”
- 16. Overall Findings Renaming is a common activity in software implementation • Most identifiers typically undergo a single rename • However, rename chains frequently occur in projects – methods are frequently associated with rename chains • Renames in a chain occur days apart – variables typically having the shortest duration (approx. two days) and attributes the longest • Renames in a rename chain are usually performed by the same developer • Multi-developer chains usually involve two developers • The grammatical structure of the initial and last name in the chain remains the same • Code Cleanup, Refactoring, and Bug Fix/Testing are the cause for rename chains – However, these topics are at a high-level due to the nature of commit messages
- 17. Key Takeaways • Part-of-speech tags are an efficient means of studying the semantic updates a name undergoes when renamed • Academia and practitioners should not limit their focus to only the words in a name • Improvements to name recommendations and appraisal techniques • These techniques should consider the historical evolution of an identifier’s name in their evaluation process • Emphasis on the importance of using high-quality names • Academia should instill in students the importance high-quality names in the source code; practitioners should incorporate naming quality into code reviews • Challenges with automated contextualization of rename chains • Current NLP techniques are not adequate for analyzing software engineering artifacts
- 18. Conclusion & Future Work • Interpreting identifier names form the backbone of any code comprehension task • Developers perform renames to correct poor-quality names • This can continue throughout the system’s lifetime • We analyze multiple renames applied to a single identifier (i.e., rename chain) • Almost all projects exhibit this phenomenon, with an average chain size of two renames • We report on characteristics such as the interval between renames, developers responsible for chain construction, grammatical changes, and motivation • Future work: Human subject study • Validate our empirical findings with developers of varying experience and skills