Traditional software defect localization aims to locate defective files, methods, or code lines based on symptoms such as defect reports. In comparison, Just-In-Time (JIT) software defect localization focuses on identifying defective code lines when a defective code change is initially submitted. It can identify issues at the code line level before the defect becomes apparent, preventing it from adversely affecting the software. Although researchers have proposed various methods for JIT defect localization, existing methods still have the following shortcomings: (1) Most methods rely heavily on tokens from single code lines to calculate naturalness for defect localization, which makes it challenging to effectively distinguish between code lines that have the same content but different labels (defective code lines or non-defective code lines) - termed Duplicate Lines with Different Labels (DLDL). (2) Existing methods represent code in the form of sequences, neglecting the structural information of the code. Therefore, we propose a JIT defect localization method based on code graph representation. First, we construct code linelevel code graphs for code changes to distinguish DLDL explicitly. Next, to extract sequential and structural information from the code, we propose a code graph representation model with contrastive learning to generate graph feature vectors and node scores with rich semantics. Finally, we calculate the naturalness of code lines based on the graph feature vectors and node scores. Using this naturalness, we identify defective code lines. Experimental results show that our JIT defect localization method outperforms the state-of-the-art methods.

Abstract Just‐In‐Time (JIT) defect prediction aims to predict the defect proneness of software changes when they are initially submitted. It has become a hot topic in software defect prediction due to its timely manner and traceability. Researchers have proposed many JIT defect prediction approaches. However, these approaches cannot effectively utilise line labels representing added or removed lines and ignore the noise caused by defect‐irrelevant files. Therefore, a JIT defect prediction model enhanced by the joint method of line label Fusion and file Filtering (JIT‐FF) is proposed. Firstly, to distinguish added and removed lines while preserving the original software changes information, the authors represent the code changes as original, added, and removed codes according to line labels. Secondly, to obtain semantics‐enhanced code representation, a cross‐attention‐based line label fusion method to perform complementary feature enhancement is proposed. Thirdly, to generate code changes containing fewer defect‐irrelevant files, the authors formalise the file filtering as a sequential decision problem and propose a reinforcement learning‐based file filtering method. Finally, based on generated code changes, CodeBERT‐based commit representation and multi‐layer perceptron‐based defect prediction are performed to identify the defective software changes. The experiments demonstrate that JIT‐FF can predict defective software changes more effectively.