Deep Learning Anti-Patterns from Code Metrics History

Abstract

Anti-patterns are poor solutions to recurring design problems. Number of empirical studies have highlighted the negative impact of anti-patterns on software maintenance which motivated the development of various detection techniques. Most of these approaches rely on structural metrics of software systems to identify affected components while others exploit historical information by analyzing co-changes occurring between code components. By relying solely on one aspect of software systems (i.e., structural or historical), existing approaches miss some precious information which limits their performances.In this paper, we propose CAME (Convolutional Analysis of code Metrics Evolution), a deep-learning based approach that relies on both structural and historical information to detect anti-patterns. Our approach exploits historical values of structural code metrics mined from version control systems and uses a Convolutional Neural Network classifier to infer the presence of anti-patterns from this information. We experiment our approach for the widely know God Class anti-pattern and evaluate its performances on three software systems. With the results of our study, we show that: (1) using historical values of source code metrics allows to increase the precision; (2) CAME outperforms existing static machine-learning classifiers; and (3) CAME outperforms existing detection tools.

Towards a systematic approach to manual annotation of code smells

“…We did not include historical indicators in our code characteristics in this iteration, as we derived our model by examining two smells, Large Class and Long Method. Though historic indicators can be used to detect Large Classes (Barbez et al, 2019), we did not find that any code smell specification, heuristics or annotator reasoning provided in the analyzed literature analyze historical factors when deciding on whether the class suffers from this smell. Furthermore, Moha et al (2010) showed that their domain analysis that did not include historic indicators is complete enough to describe a whole range of smells.…”

Automatic detection of Long Method and God Class code smells through neural source code embeddings

“…We also looked for the tools that extract the methods through static analysis (i.e., they do not require compiled code) to avoid compilation issues. We selected the CK Tool (Aniche, 2015) and RepositoryMiner (Barbez et al, 2019) for metric extraction based on these requirements.…”

“…Such embeddings are helpful for the classification of security-relevant commits (Sabetta & Bezzi, 2018;Lozoya et al, 2021), log message generation, bug fixing patch identification, and just-in-time defect prediction (Hoang et al, 2020). We did not consider this group of embeddings for code smell detection, although code change history can be helpful for God Class detection (Barbez et al, 2019;Palomba et al, 2014). The MLCQ dataset contains samples from 792 projects and extracting the needed embeddings would be extremely time intensive.…”

Automatska Detekcija Indikatora Loše Dizajniranog Koda Bazirana Na Istoriji Promena Koda

“…Obeležja koja se ekstrahuju iz koda su strukturne metrike i istorijske metrike (vrednosti strukturnih metrika u n prethodnih commit-ova), što predstavlja sličnost sa pristupom opisanim u ovom radu. Autori rada [7] predstavili su CAME 2 (Convolutional Analysis of code Metrics Evolution), pristup zasnovan na dubokom učenju (deep-learning) koji se oslanja na strukturne i istorijske vrednosti metrika u cilju detektovanja code smell-ova. Za merenje performansi je upotrebljena F-mera, kao i u ovom radu.…”

“…U radu je rečeno da je prosečna dužina sekvence između 3 i 5. U svim navedenim radovima, sem u radu [7] gde je korišćen cross-project 4 , korišćen je within-project 5 pristup evaluacije modela. Pored toga, u svim radovima je korišćena F-mera za evaluaciju modela.…”