Enhancing the Unified Features to Locate Buggy Files by Exploiting the Sequential Nature of Source Code

Abstract

Bug reports provide an effective way for end-users to disclose potential bugs hidden in a software system, while automatically locating the potential buggy source files according to a bug report remains a great challenge in software maintenance. Many previous approaches represent bug reports and source code from lexical and structural information correlated their relevance by measuring their similarity, and recently a CNN-based model is proposed to learn the unified features for bug localization, which overcomes the difficulty in modeling natural and programming languages with different structural semantics. However, previous studies fail to capture the sequential nature of source code, which carries additional semantics beyond the lexical and structural terms and such information is vital in modeling program functionalities and behaviors. In this paper, we propose a novel model LS-CNN, which enhances the unified features by exploiting the sequential nature of source code. LS-CNN combines CNN and LSTM to extract semantic features for automatically identifying potential buggy source code according to a bug report. Experimental results on widely-used software projects indicate that LS-CNN significantly outperforms the state-of-the-art methods in locating buggy files.

PatchNet: Hierarchical Deep Learning-Based Stable Patch Identification for the Linux Kernel

“…The embedding vectors are concatenated and then used to compute a prediction score for the patch. Different from existing deep learning techniques working on the source code [16], [17], [24], [36], [44], [66], [68], our hierarchical deep learning-based architecture takes into account the structure of code changes (i.e., files, hunks, lines) and the sequential nature of source code (by considering each line of code as a sequence of words) to predict stable patches in the Linux kernel.…”

“…In Tian et al's work, stable kernels were considered as a source of bugfixing patches in the training and testing data. • LS-CNN: Huo et al [24] combined LSTM [23] and CNN [39] to localize potential buggy source files based on bug report information. They used CNN to learn a representation of the bug report and a combination of LSTM and CNN to learn the structure of the code.…”

“…Learning of both code and text representations. Huo and Li proposed a model, LS-CNN, for classifying if a source code file is related to a bug report (i.e., the source code file needs to be fixed to resolve the bug report) [24]. LS-CNN is the first code representation learning method that combines CNN and LSTM (a specific type of RNN) to extract semantic representations of both code (in their case: a source code file) and text (in their case: a bug report).…”

“…The first patch changes various return values of the function csum_tree_block. The commit message is on lines 1-10 and the code changes are on lines [11][12][13][14][15][16][17][18][19][20][21][22][23][24][25]. The code changes consist of multiple hunks, only the first of which is shown in detail (lines [15][16][17][18][19][20][21][22][23].…”

“…The key challenge is to accurately represent the structure of code changes, which are not contiguous text like the commit message, but rather amount to scattered fragments of removed and added code across multiple files, within multiple hunks. Thus, different from existing deep learning techniques working on source code [24], [36], [66], [68], PatchNet constructs separate embedding vectors representing the removed code and the added code in each hunk of each affected file in the given patch. The information about a file's hunks are then concatenated to build an embedding vector for the affected file.…”