An Empirical Study of Code Smells in Transformer-based Code Generation Techniques

• Published in: Proceedings / IEEE International Working Conference on Source Code Analysis and Manipulation pp. 71 - 82
• Main Authors: Siddiq, Mohammed Latif, Majumder, Shafayat H., Mim, Maisha R., Jajodia, Sourov, Santos, Joanna C. S.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.10.2022
• Subjects: Analytical models; code generation; code smell; Codes; GitHub copilot; pre-trained model; Security; security smell; Source coding; Training; transformer; Transformers; Writing

Prior works have developed transformer-based language learning models to automatically generate source code for a task without compilation errors. The datasets used to train these techniques include samples from open source projects which may not be free of security flaws, code smells, and violation...